v8/src/arm/lithium-codegen-arm.cc

5946 lines
201 KiB
C++
Raw Normal View History

// Copyright 2012 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "src/v8.h"
#include "src/arm/lithium-codegen-arm.h"
#include "src/arm/lithium-gap-resolver-arm.h"
#include "src/base/bits.h"
#include "src/code-factory.h"
#include "src/code-stubs.h"
#include "src/cpu-profiler.h"
#include "src/hydrogen-osr.h"
#include "src/ic/ic.h"
#include "src/ic/stub-cache.h"
namespace v8 {
namespace internal {
class SafepointGenerator FINAL : public CallWrapper {
public:
SafepointGenerator(LCodeGen* codegen,
LPointerMap* pointers,
Safepoint::DeoptMode mode)
: codegen_(codegen),
pointers_(pointers),
deopt_mode_(mode) { }
virtual ~SafepointGenerator() {}
void BeforeCall(int call_size) const OVERRIDE {}
void AfterCall() const OVERRIDE {
codegen_->RecordSafepoint(pointers_, deopt_mode_);
}
private:
LCodeGen* codegen_;
LPointerMap* pointers_;
Safepoint::DeoptMode deopt_mode_;
};
#define __ masm()->
bool LCodeGen::GenerateCode() {
LPhase phase("Z_Code generation", chunk());
DCHECK(is_unused());
status_ = GENERATING;
// Open a frame scope to indicate that there is a frame on the stack. The
// NONE indicates that the scope shouldn't actually generate code to set up
// the frame (that is done in GeneratePrologue).
FrameScope frame_scope(masm_, StackFrame::NONE);
return GeneratePrologue() && GenerateBody() && GenerateDeferredCode() &&
GenerateJumpTable() && GenerateSafepointTable();
}
void LCodeGen::FinishCode(Handle<Code> code) {
DCHECK(is_done());
code->set_stack_slots(GetStackSlotCount());
code->set_safepoint_table_offset(safepoints_.GetCodeOffset());
PopulateDeoptimizationData(code);
}
void LCodeGen::SaveCallerDoubles() {
DCHECK(info()->saves_caller_doubles());
DCHECK(NeedsEagerFrame());
Comment(";;; Save clobbered callee double registers");
int count = 0;
BitVector* doubles = chunk()->allocated_double_registers();
BitVector::Iterator save_iterator(doubles);
while (!save_iterator.Done()) {
__ vstr(DwVfpRegister::FromAllocationIndex(save_iterator.Current()),
MemOperand(sp, count * kDoubleSize));
save_iterator.Advance();
count++;
}
}
void LCodeGen::RestoreCallerDoubles() {
DCHECK(info()->saves_caller_doubles());
DCHECK(NeedsEagerFrame());
Comment(";;; Restore clobbered callee double registers");
BitVector* doubles = chunk()->allocated_double_registers();
BitVector::Iterator save_iterator(doubles);
int count = 0;
while (!save_iterator.Done()) {
__ vldr(DwVfpRegister::FromAllocationIndex(save_iterator.Current()),
MemOperand(sp, count * kDoubleSize));
save_iterator.Advance();
count++;
}
}
bool LCodeGen::GeneratePrologue() {
DCHECK(is_generating());
if (info()->IsOptimizing()) {
ProfileEntryHookStub::MaybeCallEntryHook(masm_);
#ifdef DEBUG
if (strlen(FLAG_stop_at) > 0 &&
info_->function()->name()->IsUtf8EqualTo(CStrVector(FLAG_stop_at))) {
__ stop("stop_at");
}
#endif
// r1: Callee's JS function.
// cp: Callee's context.
// pp: Callee's constant pool pointer (if FLAG_enable_ool_constant_pool)
// fp: Caller's frame pointer.
// lr: Caller's pc.
// Sloppy mode functions and builtins need to replace the receiver with the
// global proxy when called as functions (without an explicit receiver
// object).
if (graph()->this_has_uses() && is_sloppy(info_->language_mode()) &&
!info_->is_native()) {
Label ok;
int receiver_offset = info_->scope()->num_parameters() * kPointerSize;
__ ldr(r2, MemOperand(sp, receiver_offset));
__ CompareRoot(r2, Heap::kUndefinedValueRootIndex);
__ b(ne, &ok);
__ ldr(r2, GlobalObjectOperand());
__ ldr(r2, FieldMemOperand(r2, GlobalObject::kGlobalProxyOffset));
__ str(r2, MemOperand(sp, receiver_offset));
__ bind(&ok);
}
}
info()->set_prologue_offset(masm_->pc_offset());
if (NeedsEagerFrame()) {
if (info()->IsStub()) {
__ StubPrologue();
} else {
__ Prologue(info()->IsCodePreAgingActive());
}
frame_is_built_ = true;
info_->AddNoFrameRange(0, masm_->pc_offset());
}
// Reserve space for the stack slots needed by the code.
int slots = GetStackSlotCount();
if (slots > 0) {
if (FLAG_debug_code) {
__ sub(sp, sp, Operand(slots * kPointerSize));
__ push(r0);
__ push(r1);
__ add(r0, sp, Operand(slots * kPointerSize));
__ mov(r1, Operand(kSlotsZapValue));
Label loop;
__ bind(&loop);
__ sub(r0, r0, Operand(kPointerSize));
__ str(r1, MemOperand(r0, 2 * kPointerSize));
__ cmp(r0, sp);
__ b(ne, &loop);
__ pop(r1);
__ pop(r0);
} else {
__ sub(sp, sp, Operand(slots * kPointerSize));
}
}
if (info()->saves_caller_doubles()) {
SaveCallerDoubles();
}
// Possibly allocate a local context.
int heap_slots = info()->num_heap_slots() - Context::MIN_CONTEXT_SLOTS;
if (heap_slots > 0) {
Comment(";;; Allocate local context");
bool need_write_barrier = true;
// Argument to NewContext is the function, which is in r1.
if (heap_slots <= FastNewContextStub::kMaximumSlots) {
FastNewContextStub stub(isolate(), heap_slots);
__ CallStub(&stub);
// Result of FastNewContextStub is always in new space.
need_write_barrier = false;
} else {
__ push(r1);
__ CallRuntime(Runtime::kNewFunctionContext, 1);
}
RecordSafepoint(Safepoint::kNoLazyDeopt);
// Context is returned in both r0 and cp. It replaces the context
// passed to us. It's saved in the stack and kept live in cp.
__ mov(cp, r0);
__ str(r0, MemOperand(fp, StandardFrameConstants::kContextOffset));
// Copy any necessary parameters into the context.
int num_parameters = scope()->num_parameters();
for (int i = 0; i < num_parameters; i++) {
Variable* var = scope()->parameter(i);
if (var->IsContextSlot()) {
int parameter_offset = StandardFrameConstants::kCallerSPOffset +
(num_parameters - 1 - i) * kPointerSize;
// Load parameter from stack.
__ ldr(r0, MemOperand(fp, parameter_offset));
// Store it in the context.
MemOperand target = ContextOperand(cp, var->index());
__ str(r0, target);
// Update the write barrier. This clobbers r3 and r0.
if (need_write_barrier) {
__ RecordWriteContextSlot(
cp,
target.offset(),
r0,
r3,
GetLinkRegisterState(),
kSaveFPRegs);
} else if (FLAG_debug_code) {
Label done;
__ JumpIfInNewSpace(cp, r0, &done);
__ Abort(kExpectedNewSpaceObject);
__ bind(&done);
}
}
}
Comment(";;; End allocate local context");
}
// Trace the call.
if (FLAG_trace && info()->IsOptimizing()) {
// We have not executed any compiled code yet, so cp still holds the
// incoming context.
__ CallRuntime(Runtime::kTraceEnter, 0);
}
return !is_aborted();
}
void LCodeGen::GenerateOsrPrologue() {
// Generate the OSR entry prologue at the first unknown OSR value, or if there
// are none, at the OSR entrypoint instruction.
if (osr_pc_offset_ >= 0) return;
osr_pc_offset_ = masm()->pc_offset();
// Adjust the frame size, subsuming the unoptimized frame into the
// optimized frame.
int slots = GetStackSlotCount() - graph()->osr()->UnoptimizedFrameSlots();
DCHECK(slots >= 0);
__ sub(sp, sp, Operand(slots * kPointerSize));
}
void LCodeGen::GenerateBodyInstructionPre(LInstruction* instr) {
if (instr->IsCall()) {
EnsureSpaceForLazyDeopt(Deoptimizer::patch_size());
}
if (!instr->IsLazyBailout() && !instr->IsGap()) {
safepoints_.BumpLastLazySafepointIndex();
}
}
bool LCodeGen::GenerateDeferredCode() {
DCHECK(is_generating());
if (deferred_.length() > 0) {
for (int i = 0; !is_aborted() && i < deferred_.length(); i++) {
LDeferredCode* code = deferred_[i];
HValue* value =
instructions_->at(code->instruction_index())->hydrogen_value();
RecordAndWritePosition(
chunk()->graph()->SourcePositionToScriptPosition(value->position()));
Comment(";;; <@%d,#%d> "
"-------------------- Deferred %s --------------------",
code->instruction_index(),
code->instr()->hydrogen_value()->id(),
code->instr()->Mnemonic());
__ bind(code->entry());
if (NeedsDeferredFrame()) {
Comment(";;; Build frame");
DCHECK(!frame_is_built_);
DCHECK(info()->IsStub());
frame_is_built_ = true;
__ PushFixedFrame();
__ mov(scratch0(), Operand(Smi::FromInt(StackFrame::STUB)));
__ push(scratch0());
__ add(fp, sp, Operand(StandardFrameConstants::kFixedFrameSizeFromFp));
Comment(";;; Deferred code");
}
code->Generate();
if (NeedsDeferredFrame()) {
Comment(";;; Destroy frame");
DCHECK(frame_is_built_);
__ pop(ip);
__ PopFixedFrame();
frame_is_built_ = false;
}
__ jmp(code->exit());
}
}
// Force constant pool emission at the end of the deferred code to make
// sure that no constant pools are emitted after.
masm()->CheckConstPool(true, false);
return !is_aborted();
}
bool LCodeGen::GenerateJumpTable() {
// Check that the jump table is accessible from everywhere in the function
// code, i.e. that offsets to the table can be encoded in the 24bit signed
// immediate of a branch instruction.
// To simplify we consider the code size from the first instruction to the
// end of the jump table. We also don't consider the pc load delta.
// Each entry in the jump table generates one instruction and inlines one
// 32bit data after it.
if (!is_int24((masm()->pc_offset() / Assembler::kInstrSize) +
jump_table_.length() * 7)) {
Abort(kGeneratedCodeIsTooLarge);
}
if (jump_table_.length() > 0) {
Label needs_frame, call_deopt_entry;
Comment(";;; -------------------- Jump table --------------------");
Address base = jump_table_[0].address;
Register entry_offset = scratch0();
int length = jump_table_.length();
for (int i = 0; i < length; i++) {
Deoptimizer::JumpTableEntry* table_entry = &jump_table_[i];
__ bind(&table_entry->label);
DCHECK_EQ(jump_table_[0].bailout_type, table_entry->bailout_type);
Address entry = table_entry->address;
DeoptComment(table_entry->deopt_info);
// Second-level deopt table entries are contiguous and small, so instead
// of loading the full, absolute address of each one, load an immediate
// offset which will be added to the base address later.
__ mov(entry_offset, Operand(entry - base));
if (table_entry->needs_frame) {
DCHECK(!info()->saves_caller_doubles());
Comment(";;; call deopt with frame");
__ PushFixedFrame();
__ bl(&needs_frame);
} else {
__ bl(&call_deopt_entry);
}
info()->LogDeoptCallPosition(masm()->pc_offset(),
table_entry->deopt_info.inlining_id);
masm()->CheckConstPool(false, false);
}
if (needs_frame.is_linked()) {
__ bind(&needs_frame);
// This variant of deopt can only be used with stubs. Since we don't
// have a function pointer to install in the stack frame that we're
// building, install a special marker there instead.
DCHECK(info()->IsStub());
__ mov(ip, Operand(Smi::FromInt(StackFrame::STUB)));
__ push(ip);
__ add(fp, sp, Operand(StandardFrameConstants::kFixedFrameSizeFromFp));
}
Comment(";;; call deopt");
__ bind(&call_deopt_entry);
if (info()->saves_caller_doubles()) {
DCHECK(info()->IsStub());
RestoreCallerDoubles();
}
// Add the base address to the offset previously loaded in entry_offset.
__ add(entry_offset, entry_offset,
Operand(ExternalReference::ForDeoptEntry(base)));
__ bx(entry_offset);
}
// Force constant pool emission at the end of the deopt jump table to make
// sure that no constant pools are emitted after.
masm()->CheckConstPool(true, false);
// The deoptimization jump table is the last part of the instruction
// sequence. Mark the generated code as done unless we bailed out.
if (!is_aborted()) status_ = DONE;
return !is_aborted();
}
bool LCodeGen::GenerateSafepointTable() {
DCHECK(is_done());
safepoints_.Emit(masm(), GetStackSlotCount());
return !is_aborted();
}
Register LCodeGen::ToRegister(int index) const {
return Register::FromAllocationIndex(index);
}
DwVfpRegister LCodeGen::ToDoubleRegister(int index) const {
return DwVfpRegister::FromAllocationIndex(index);
}
Register LCodeGen::ToRegister(LOperand* op) const {
DCHECK(op->IsRegister());
return ToRegister(op->index());
}
Register LCodeGen::EmitLoadRegister(LOperand* op, Register scratch) {
if (op->IsRegister()) {
return ToRegister(op->index());
} else if (op->IsConstantOperand()) {
LConstantOperand* const_op = LConstantOperand::cast(op);
HConstant* constant = chunk_->LookupConstant(const_op);
Handle<Object> literal = constant->handle(isolate());
Representation r = chunk_->LookupLiteralRepresentation(const_op);
if (r.IsInteger32()) {
DCHECK(literal->IsNumber());
__ mov(scratch, Operand(static_cast<int32_t>(literal->Number())));
} else if (r.IsDouble()) {
Abort(kEmitLoadRegisterUnsupportedDoubleImmediate);
} else {
DCHECK(r.IsSmiOrTagged());
__ Move(scratch, literal);
}
return scratch;
} else if (op->IsStackSlot()) {
__ ldr(scratch, ToMemOperand(op));
return scratch;
}
UNREACHABLE();
return scratch;
}
DwVfpRegister LCodeGen::ToDoubleRegister(LOperand* op) const {
DCHECK(op->IsDoubleRegister());
return ToDoubleRegister(op->index());
}
DwVfpRegister LCodeGen::EmitLoadDoubleRegister(LOperand* op,
SwVfpRegister flt_scratch,
DwVfpRegister dbl_scratch) {
if (op->IsDoubleRegister()) {
return ToDoubleRegister(op->index());
} else if (op->IsConstantOperand()) {
LConstantOperand* const_op = LConstantOperand::cast(op);
HConstant* constant = chunk_->LookupConstant(const_op);
Handle<Object> literal = constant->handle(isolate());
Representation r = chunk_->LookupLiteralRepresentation(const_op);
if (r.IsInteger32()) {
DCHECK(literal->IsNumber());
__ mov(ip, Operand(static_cast<int32_t>(literal->Number())));
__ vmov(flt_scratch, ip);
__ vcvt_f64_s32(dbl_scratch, flt_scratch);
return dbl_scratch;
} else if (r.IsDouble()) {
Abort(kUnsupportedDoubleImmediate);
} else if (r.IsTagged()) {
Abort(kUnsupportedTaggedImmediate);
}
} else if (op->IsStackSlot()) {
// TODO(regis): Why is vldr not taking a MemOperand?
// __ vldr(dbl_scratch, ToMemOperand(op));
MemOperand mem_op = ToMemOperand(op);
__ vldr(dbl_scratch, mem_op.rn(), mem_op.offset());
return dbl_scratch;
}
UNREACHABLE();
return dbl_scratch;
}
Handle<Object> LCodeGen::ToHandle(LConstantOperand* op) const {
HConstant* constant = chunk_->LookupConstant(op);
DCHECK(chunk_->LookupLiteralRepresentation(op).IsSmiOrTagged());
return constant->handle(isolate());
}
bool LCodeGen::IsInteger32(LConstantOperand* op) const {
return chunk_->LookupLiteralRepresentation(op).IsSmiOrInteger32();
}
bool LCodeGen::IsSmi(LConstantOperand* op) const {
return chunk_->LookupLiteralRepresentation(op).IsSmi();
}
int32_t LCodeGen::ToInteger32(LConstantOperand* op) const {
return ToRepresentation(op, Representation::Integer32());
}
int32_t LCodeGen::ToRepresentation(LConstantOperand* op,
const Representation& r) const {
HConstant* constant = chunk_->LookupConstant(op);
int32_t value = constant->Integer32Value();
if (r.IsInteger32()) return value;
DCHECK(r.IsSmiOrTagged());
return reinterpret_cast<int32_t>(Smi::FromInt(value));
}
Smi* LCodeGen::ToSmi(LConstantOperand* op) const {
HConstant* constant = chunk_->LookupConstant(op);
return Smi::FromInt(constant->Integer32Value());
}
double LCodeGen::ToDouble(LConstantOperand* op) const {
HConstant* constant = chunk_->LookupConstant(op);
DCHECK(constant->HasDoubleValue());
return constant->DoubleValue();
}
Operand LCodeGen::ToOperand(LOperand* op) {
if (op->IsConstantOperand()) {
LConstantOperand* const_op = LConstantOperand::cast(op);
HConstant* constant = chunk()->LookupConstant(const_op);
Representation r = chunk_->LookupLiteralRepresentation(const_op);
if (r.IsSmi()) {
DCHECK(constant->HasSmiValue());
return Operand(Smi::FromInt(constant->Integer32Value()));
} else if (r.IsInteger32()) {
DCHECK(constant->HasInteger32Value());
return Operand(constant->Integer32Value());
} else if (r.IsDouble()) {
Abort(kToOperandUnsupportedDoubleImmediate);
}
DCHECK(r.IsTagged());
return Operand(constant->handle(isolate()));
} else if (op->IsRegister()) {
return Operand(ToRegister(op));
} else if (op->IsDoubleRegister()) {
Abort(kToOperandIsDoubleRegisterUnimplemented);
return Operand::Zero();
}
// Stack slots not implemented, use ToMemOperand instead.
UNREACHABLE();
return Operand::Zero();
}
static int ArgumentsOffsetWithoutFrame(int index) {
DCHECK(index < 0);
return -(index + 1) * kPointerSize;
}
MemOperand LCodeGen::ToMemOperand(LOperand* op) const {
DCHECK(!op->IsRegister());
DCHECK(!op->IsDoubleRegister());
DCHECK(op->IsStackSlot() || op->IsDoubleStackSlot());
if (NeedsEagerFrame()) {
return MemOperand(fp, StackSlotOffset(op->index()));
} else {
// Retrieve parameter without eager stack-frame relative to the
// stack-pointer.
return MemOperand(sp, ArgumentsOffsetWithoutFrame(op->index()));
}
}
MemOperand LCodeGen::ToHighMemOperand(LOperand* op) const {
DCHECK(op->IsDoubleStackSlot());
if (NeedsEagerFrame()) {
return MemOperand(fp, StackSlotOffset(op->index()) + kPointerSize);
} else {
// Retrieve parameter without eager stack-frame relative to the
// stack-pointer.
return MemOperand(
sp, ArgumentsOffsetWithoutFrame(op->index()) + kPointerSize);
}
}
void LCodeGen::WriteTranslation(LEnvironment* environment,
Translation* translation) {
if (environment == NULL) return;
// The translation includes one command per value in the environment.
int translation_size = environment->translation_size();
// The output frame height does not include the parameters.
int height = translation_size - environment->parameter_count();
WriteTranslation(environment->outer(), translation);
bool has_closure_id = !info()->closure().is_null() &&
!info()->closure().is_identical_to(environment->closure());
int closure_id = has_closure_id
? DefineDeoptimizationLiteral(environment->closure())
: Translation::kSelfLiteralId;
switch (environment->frame_type()) {
case JS_FUNCTION:
translation->BeginJSFrame(environment->ast_id(), closure_id, height);
break;
case JS_CONSTRUCT:
translation->BeginConstructStubFrame(closure_id, translation_size);
break;
case JS_GETTER:
DCHECK(translation_size == 1);
DCHECK(height == 0);
translation->BeginGetterStubFrame(closure_id);
break;
case JS_SETTER:
DCHECK(translation_size == 2);
DCHECK(height == 0);
translation->BeginSetterStubFrame(closure_id);
break;
case STUB:
translation->BeginCompiledStubFrame();
break;
case ARGUMENTS_ADAPTOR:
translation->BeginArgumentsAdaptorFrame(closure_id, translation_size);
break;
}
int object_index = 0;
int dematerialized_index = 0;
for (int i = 0; i < translation_size; ++i) {
LOperand* value = environment->values()->at(i);
AddToTranslation(environment,
translation,
value,
environment->HasTaggedValueAt(i),
environment->HasUint32ValueAt(i),
&object_index,
&dematerialized_index);
}
}
void LCodeGen::AddToTranslation(LEnvironment* environment,
Translation* translation,
LOperand* op,
bool is_tagged,
bool is_uint32,
int* object_index_pointer,
int* dematerialized_index_pointer) {
if (op == LEnvironment::materialization_marker()) {
int object_index = (*object_index_pointer)++;
if (environment->ObjectIsDuplicateAt(object_index)) {
int dupe_of = environment->ObjectDuplicateOfAt(object_index);
translation->DuplicateObject(dupe_of);
return;
}
int object_length = environment->ObjectLengthAt(object_index);
if (environment->ObjectIsArgumentsAt(object_index)) {
translation->BeginArgumentsObject(object_length);
} else {
translation->BeginCapturedObject(object_length);
}
int dematerialized_index = *dematerialized_index_pointer;
int env_offset = environment->translation_size() + dematerialized_index;
*dematerialized_index_pointer += object_length;
for (int i = 0; i < object_length; ++i) {
LOperand* value = environment->values()->at(env_offset + i);
AddToTranslation(environment,
translation,
value,
environment->HasTaggedValueAt(env_offset + i),
environment->HasUint32ValueAt(env_offset + i),
object_index_pointer,
dematerialized_index_pointer);
}
return;
}
if (op->IsStackSlot()) {
if (is_tagged) {
translation->StoreStackSlot(op->index());
} else if (is_uint32) {
translation->StoreUint32StackSlot(op->index());
} else {
translation->StoreInt32StackSlot(op->index());
}
} else if (op->IsDoubleStackSlot()) {
translation->StoreDoubleStackSlot(op->index());
} else if (op->IsRegister()) {
Register reg = ToRegister(op);
if (is_tagged) {
translation->StoreRegister(reg);
} else if (is_uint32) {
translation->StoreUint32Register(reg);
} else {
translation->StoreInt32Register(reg);
}
} else if (op->IsDoubleRegister()) {
DoubleRegister reg = ToDoubleRegister(op);
translation->StoreDoubleRegister(reg);
} else if (op->IsConstantOperand()) {
HConstant* constant = chunk()->LookupConstant(LConstantOperand::cast(op));
int src_index = DefineDeoptimizationLiteral(constant->handle(isolate()));
translation->StoreLiteral(src_index);
} else {
UNREACHABLE();
}
}
int LCodeGen::CallCodeSize(Handle<Code> code, RelocInfo::Mode mode) {
int size = masm()->CallSize(code, mode);
if (code->kind() == Code::BINARY_OP_IC ||
code->kind() == Code::COMPARE_IC) {
size += Assembler::kInstrSize; // extra nop() added in CallCodeGeneric.
}
return size;
}
void LCodeGen::CallCode(Handle<Code> code,
RelocInfo::Mode mode,
LInstruction* instr,
TargetAddressStorageMode storage_mode) {
CallCodeGeneric(code, mode, instr, RECORD_SIMPLE_SAFEPOINT, storage_mode);
}
void LCodeGen::CallCodeGeneric(Handle<Code> code,
RelocInfo::Mode mode,
LInstruction* instr,
SafepointMode safepoint_mode,
TargetAddressStorageMode storage_mode) {
DCHECK(instr != NULL);
// Block literal pool emission to ensure nop indicating no inlined smi code
// is in the correct position.
Assembler::BlockConstPoolScope block_const_pool(masm());
__ Call(code, mode, TypeFeedbackId::None(), al, storage_mode);
RecordSafepointWithLazyDeopt(instr, safepoint_mode);
Avoid patching code after the call to binary operation stub in optimized code This patch just adds a nop after the call to the binary operation stub in optimized code to avoid the patching for the inlined smi case used in the full code generator to kick in if the next instruction generated by the lithium code generator should accidentially enable that. For calls generated by CallCodeGeneric this was already handled on Intel platforms, but missing on ARM. On IA-32 I did also try to check for whether the code containing the call was optimized (patch below), but that caused regressions on some benchmarks. diff --git src/ia32/ic-ia32.cc src/ia32/ic-ia32.cc index 5f143b1..f70e208 100644 --- src/ia32/ic-ia32.cc +++ src/ia32/ic-ia32.cc @@ -1603,12 +1603,18 @@ void CompareIC::UpdateCaches(Handle<Object> x, Handle<Object> y) { // Activate inlined smi code. if (previous_state == UNINITIALIZED) { - PatchInlinedSmiCode(address()); + PatchInlinedSmiCode(address(), isolate()); } } -void PatchInlinedSmiCode(Address address) { +void PatchInlinedSmiCode(Address address, Isolate* isolate) { + // Never patch in optimized code. + Code* code = isolate->pc_to_code_cache()->GetCacheEntry(address)->code; + if (code->kind() == Code::OPTIMIZED_FUNCTION) { + return; + } + // The address of the instruction following the call. Address test_instruction_address = address + Assembler::kCallTargetAddressOffset; diff --git src/ic.cc src/ic.cc index f70f75a..62e79da 100644 --- src/ic.cc +++ src/ic.cc @@ -2384,7 +2384,7 @@ RUNTIME_FUNCTION(MaybeObject*, BinaryOp_Patch) { // Activate inlined smi code. if (previous_type == BinaryOpIC::UNINITIALIZED) { - PatchInlinedSmiCode(ic.address()); + PatchInlinedSmiCode(ic.address(), isolate); } } diff --git src/ic.h src/ic.h index 11c2e3a..9ef4b20 100644 --- src/ic.h +++ src/ic.h @@ -721,7 +721,7 @@ class CompareIC: public IC { }; // Helper for BinaryOpIC and CompareIC. -void PatchInlinedSmiCode(Address address); +void PatchInlinedSmiCode(Address address, Isolate* isolate); } } // namespace v8::internal R=danno@chromium.org BUG=none TEST=none Review URL: http://codereview.chromium.org//7350015 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8623 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2011-07-13 09:31:17 +00:00
// Signal that we don't inline smi code before these stubs in the
// optimizing code generator.
if (code->kind() == Code::BINARY_OP_IC ||
code->kind() == Code::COMPARE_IC) {
__ nop();
}
}
void LCodeGen::CallRuntime(const Runtime::Function* function,
int num_arguments,
LInstruction* instr,
SaveFPRegsMode save_doubles) {
DCHECK(instr != NULL);
__ CallRuntime(function, num_arguments, save_doubles);
RecordSafepointWithLazyDeopt(instr, RECORD_SIMPLE_SAFEPOINT);
}
void LCodeGen::LoadContextFromDeferred(LOperand* context) {
if (context->IsRegister()) {
__ Move(cp, ToRegister(context));
} else if (context->IsStackSlot()) {
__ ldr(cp, ToMemOperand(context));
} else if (context->IsConstantOperand()) {
HConstant* constant =
chunk_->LookupConstant(LConstantOperand::cast(context));
__ Move(cp, Handle<Object>::cast(constant->handle(isolate())));
} else {
UNREACHABLE();
}
}
void LCodeGen::CallRuntimeFromDeferred(Runtime::FunctionId id,
int argc,
LInstruction* instr,
LOperand* context) {
LoadContextFromDeferred(context);
__ CallRuntimeSaveDoubles(id);
RecordSafepointWithRegisters(
instr->pointer_map(), argc, Safepoint::kNoLazyDeopt);
}
void LCodeGen::RegisterEnvironmentForDeoptimization(LEnvironment* environment,
Safepoint::DeoptMode mode) {
environment->set_has_been_used();
if (!environment->HasBeenRegistered()) {
// Physical stack frame layout:
// -x ............. -4 0 ..................................... y
// [incoming arguments] [spill slots] [pushed outgoing arguments]
// Layout of the environment:
// 0 ..................................................... size-1
// [parameters] [locals] [expression stack including arguments]
// Layout of the translation:
// 0 ........................................................ size - 1 + 4
// [expression stack including arguments] [locals] [4 words] [parameters]
// |>------------ translation_size ------------<|
int frame_count = 0;
int jsframe_count = 0;
for (LEnvironment* e = environment; e != NULL; e = e->outer()) {
++frame_count;
if (e->frame_type() == JS_FUNCTION) {
++jsframe_count;
}
}
Translation translation(&translations_, frame_count, jsframe_count, zone());
WriteTranslation(environment, &translation);
int deoptimization_index = deoptimizations_.length();
int pc_offset = masm()->pc_offset();
environment->Register(deoptimization_index,
translation.index(),
(mode == Safepoint::kLazyDeopt) ? pc_offset : -1);
deoptimizations_.Add(environment, zone());
}
}
void LCodeGen::DeoptimizeIf(Condition condition, LInstruction* instr,
Deoptimizer::DeoptReason deopt_reason,
Deoptimizer::BailoutType bailout_type) {
LEnvironment* environment = instr->environment();
RegisterEnvironmentForDeoptimization(environment, Safepoint::kNoLazyDeopt);
DCHECK(environment->HasBeenRegistered());
int id = environment->deoptimization_index();
DCHECK(info()->IsOptimizing() || info()->IsStub());
Address entry =
Deoptimizer::GetDeoptimizationEntry(isolate(), id, bailout_type);
if (entry == NULL) {
Abort(kBailoutWasNotPrepared);
return;
}
if (FLAG_deopt_every_n_times != 0 && !info()->IsStub()) {
Register scratch = scratch0();
ExternalReference count = ExternalReference::stress_deopt_count(isolate());
// Store the condition on the stack if necessary
if (condition != al) {
__ mov(scratch, Operand::Zero(), LeaveCC, NegateCondition(condition));
__ mov(scratch, Operand(1), LeaveCC, condition);
__ push(scratch);
}
__ push(r1);
__ mov(scratch, Operand(count));
__ ldr(r1, MemOperand(scratch));
__ sub(r1, r1, Operand(1), SetCC);
__ mov(r1, Operand(FLAG_deopt_every_n_times), LeaveCC, eq);
__ str(r1, MemOperand(scratch));
__ pop(r1);
if (condition != al) {
// Clean up the stack before the deoptimizer call
__ pop(scratch);
}
__ Call(entry, RelocInfo::RUNTIME_ENTRY, eq);
// 'Restore' the condition in a slightly hacky way. (It would be better
// to use 'msr' and 'mrs' instructions here, but they are not supported by
// our ARM simulator).
if (condition != al) {
condition = ne;
__ cmp(scratch, Operand::Zero());
}
}
if (info()->ShouldTrapOnDeopt()) {
__ stop("trap_on_deopt", condition);
}
Deoptimizer::DeoptInfo deopt_info = MakeDeoptInfo(instr, deopt_reason);
DCHECK(info()->IsStub() || frame_is_built_);
// Go through jump table if we need to handle condition, build frame, or
// restore caller doubles.
if (condition == al && frame_is_built_ &&
!info()->saves_caller_doubles()) {
DeoptComment(deopt_info);
__ Call(entry, RelocInfo::RUNTIME_ENTRY);
info()->LogDeoptCallPosition(masm()->pc_offset(), deopt_info.inlining_id);
} else {
Deoptimizer::JumpTableEntry table_entry(entry, deopt_info, bailout_type,
!frame_is_built_);
// We often have several deopts to the same entry, reuse the last
// jump entry if this is the case.
if (FLAG_trace_deopt || isolate()->cpu_profiler()->is_profiling() ||
jump_table_.is_empty() ||
!table_entry.IsEquivalentTo(jump_table_.last())) {
jump_table_.Add(table_entry, zone());
}
__ b(condition, &jump_table_.last().label);
}
}
void LCodeGen::DeoptimizeIf(Condition condition, LInstruction* instr,
Deoptimizer::DeoptReason deopt_reason) {
Deoptimizer::BailoutType bailout_type = info()->IsStub()
? Deoptimizer::LAZY
: Deoptimizer::EAGER;
DeoptimizeIf(condition, instr, deopt_reason, bailout_type);
}
void LCodeGen::PopulateDeoptimizationData(Handle<Code> code) {
int length = deoptimizations_.length();
if (length == 0) return;
Handle<DeoptimizationInputData> data =
DeoptimizationInputData::New(isolate(), length, TENURED);
Handle<ByteArray> translations =
translations_.CreateByteArray(isolate()->factory());
data->SetTranslationByteArray(*translations);
data->SetInlinedFunctionCount(Smi::FromInt(inlined_function_count_));
data->SetOptimizationId(Smi::FromInt(info_->optimization_id()));
if (info_->IsOptimizing()) {
// Reference to shared function info does not change between phases.
AllowDeferredHandleDereference allow_handle_dereference;
data->SetSharedFunctionInfo(*info_->shared_info());
} else {
data->SetSharedFunctionInfo(Smi::FromInt(0));
}
data->SetWeakCellCache(Smi::FromInt(0));
Handle<FixedArray> literals =
factory()->NewFixedArray(deoptimization_literals_.length(), TENURED);
{ AllowDeferredHandleDereference copy_handles;
for (int i = 0; i < deoptimization_literals_.length(); i++) {
literals->set(i, *deoptimization_literals_[i]);
}
data->SetLiteralArray(*literals);
}
data->SetOsrAstId(Smi::FromInt(info_->osr_ast_id().ToInt()));
data->SetOsrPcOffset(Smi::FromInt(osr_pc_offset_));
// Populate the deoptimization entries.
for (int i = 0; i < length; i++) {
LEnvironment* env = deoptimizations_[i];
data->SetAstId(i, env->ast_id());
data->SetTranslationIndex(i, Smi::FromInt(env->translation_index()));
data->SetArgumentsStackHeight(i,
Smi::FromInt(env->arguments_stack_height()));
data->SetPc(i, Smi::FromInt(env->pc_offset()));
}
code->set_deoptimization_data(*data);
}
int LCodeGen::DefineDeoptimizationLiteral(Handle<Object> literal) {
int result = deoptimization_literals_.length();
for (int i = 0; i < deoptimization_literals_.length(); ++i) {
if (deoptimization_literals_[i].is_identical_to(literal)) return i;
}
deoptimization_literals_.Add(literal, zone());
return result;
}
void LCodeGen::PopulateDeoptimizationLiteralsWithInlinedFunctions() {
DCHECK(deoptimization_literals_.length() == 0);
const ZoneList<Handle<JSFunction> >* inlined_closures =
chunk()->inlined_closures();
for (int i = 0, length = inlined_closures->length();
i < length;
i++) {
DefineDeoptimizationLiteral(inlined_closures->at(i));
}
inlined_function_count_ = deoptimization_literals_.length();
}
void LCodeGen::RecordSafepointWithLazyDeopt(
LInstruction* instr, SafepointMode safepoint_mode) {
if (safepoint_mode == RECORD_SIMPLE_SAFEPOINT) {
RecordSafepoint(instr->pointer_map(), Safepoint::kLazyDeopt);
} else {
DCHECK(safepoint_mode == RECORD_SAFEPOINT_WITH_REGISTERS_AND_NO_ARGUMENTS);
RecordSafepointWithRegisters(
instr->pointer_map(), 0, Safepoint::kLazyDeopt);
}
}
void LCodeGen::RecordSafepoint(
LPointerMap* pointers,
Safepoint::Kind kind,
int arguments,
Safepoint::DeoptMode deopt_mode) {
DCHECK(expected_safepoint_kind_ == kind);
const ZoneList<LOperand*>* operands = pointers->GetNormalizedOperands();
Safepoint safepoint = safepoints_.DefineSafepoint(masm(),
kind, arguments, deopt_mode);
for (int i = 0; i < operands->length(); i++) {
LOperand* pointer = operands->at(i);
if (pointer->IsStackSlot()) {
safepoint.DefinePointerSlot(pointer->index(), zone());
} else if (pointer->IsRegister() && (kind & Safepoint::kWithRegisters)) {
safepoint.DefinePointerRegister(ToRegister(pointer), zone());
}
}
if (FLAG_enable_ool_constant_pool && (kind & Safepoint::kWithRegisters)) {
// Register pp always contains a pointer to the constant pool.
safepoint.DefinePointerRegister(pp, zone());
}
}
void LCodeGen::RecordSafepoint(LPointerMap* pointers,
Safepoint::DeoptMode deopt_mode) {
RecordSafepoint(pointers, Safepoint::kSimple, 0, deopt_mode);
}
void LCodeGen::RecordSafepoint(Safepoint::DeoptMode deopt_mode) {
LPointerMap empty_pointers(zone());
RecordSafepoint(&empty_pointers, deopt_mode);
}
void LCodeGen::RecordSafepointWithRegisters(LPointerMap* pointers,
int arguments,
Safepoint::DeoptMode deopt_mode) {
RecordSafepoint(
pointers, Safepoint::kWithRegisters, arguments, deopt_mode);
}
void LCodeGen::RecordAndWritePosition(int position) {
if (position == RelocInfo::kNoPosition) return;
masm()->positions_recorder()->RecordPosition(position);
masm()->positions_recorder()->WriteRecordedPositions();
}
static const char* LabelType(LLabel* label) {
if (label->is_loop_header()) return " (loop header)";
if (label->is_osr_entry()) return " (OSR entry)";
return "";
}
void LCodeGen::DoLabel(LLabel* label) {
Comment(";;; <@%d,#%d> -------------------- B%d%s --------------------",
current_instruction_,
label->hydrogen_value()->id(),
label->block_id(),
LabelType(label));
__ bind(label->label());
current_block_ = label->block_id();
DoGap(label);
}
void LCodeGen::DoParallelMove(LParallelMove* move) {
resolver_.Resolve(move);
}
void LCodeGen::DoGap(LGap* gap) {
for (int i = LGap::FIRST_INNER_POSITION;
i <= LGap::LAST_INNER_POSITION;
i++) {
LGap::InnerPosition inner_pos = static_cast<LGap::InnerPosition>(i);
LParallelMove* move = gap->GetParallelMove(inner_pos);
if (move != NULL) DoParallelMove(move);
}
}
void LCodeGen::DoInstructionGap(LInstructionGap* instr) {
DoGap(instr);
}
void LCodeGen::DoParameter(LParameter* instr) {
// Nothing to do.
}
void LCodeGen::DoCallStub(LCallStub* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->result()).is(r0));
switch (instr->hydrogen()->major_key()) {
case CodeStub::RegExpExec: {
RegExpExecStub stub(isolate());
CallCode(stub.GetCode(), RelocInfo::CODE_TARGET, instr);
break;
}
case CodeStub::SubString: {
SubStringStub stub(isolate());
CallCode(stub.GetCode(), RelocInfo::CODE_TARGET, instr);
break;
}
case CodeStub::StringCompare: {
StringCompareStub stub(isolate());
CallCode(stub.GetCode(), RelocInfo::CODE_TARGET, instr);
break;
}
default:
UNREACHABLE();
}
}
void LCodeGen::DoUnknownOSRValue(LUnknownOSRValue* instr) {
GenerateOsrPrologue();
}
void LCodeGen::DoModByPowerOf2I(LModByPowerOf2I* instr) {
Register dividend = ToRegister(instr->dividend());
int32_t divisor = instr->divisor();
DCHECK(dividend.is(ToRegister(instr->result())));
// Theoretically, a variation of the branch-free code for integer division by
// a power of 2 (calculating the remainder via an additional multiplication
// (which gets simplified to an 'and') and subtraction) should be faster, and
// this is exactly what GCC and clang emit. Nevertheless, benchmarks seem to
// indicate that positive dividends are heavily favored, so the branching
// version performs better.
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
HMod* hmod = instr->hydrogen();
int32_t mask = divisor < 0 ? -(divisor + 1) : (divisor - 1);
Label dividend_is_not_negative, done;
if (hmod->CheckFlag(HValue::kLeftCanBeNegative)) {
__ cmp(dividend, Operand::Zero());
__ b(pl, &dividend_is_not_negative);
// Note that this is correct even for kMinInt operands.
__ rsb(dividend, dividend, Operand::Zero());
__ and_(dividend, dividend, Operand(mask));
__ rsb(dividend, dividend, Operand::Zero(), SetCC);
if (hmod->CheckFlag(HValue::kBailoutOnMinusZero)) {
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
}
__ b(&done);
}
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ bind(&dividend_is_not_negative);
__ and_(dividend, dividend, Operand(mask));
__ bind(&done);
}
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
void LCodeGen::DoModByConstI(LModByConstI* instr) {
Register dividend = ToRegister(instr->dividend());
int32_t divisor = instr->divisor();
Register result = ToRegister(instr->result());
DCHECK(!dividend.is(result));
if (divisor == 0) {
DeoptimizeIf(al, instr, Deoptimizer::kDivisionByZero);
return;
}
__ TruncatingDiv(result, dividend, Abs(divisor));
__ mov(ip, Operand(Abs(divisor)));
__ smull(result, ip, result, ip);
__ sub(result, dividend, result, SetCC);
// Check for negative zero.
HMod* hmod = instr->hydrogen();
if (hmod->CheckFlag(HValue::kBailoutOnMinusZero)) {
Label remainder_not_zero;
__ b(ne, &remainder_not_zero);
__ cmp(dividend, Operand::Zero());
DeoptimizeIf(lt, instr, Deoptimizer::kMinusZero);
__ bind(&remainder_not_zero);
}
}
void LCodeGen::DoModI(LModI* instr) {
HMod* hmod = instr->hydrogen();
if (CpuFeatures::IsSupported(SUDIV)) {
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
CpuFeatureScope scope(masm(), SUDIV);
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
Register left_reg = ToRegister(instr->left());
Register right_reg = ToRegister(instr->right());
Register result_reg = ToRegister(instr->result());
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
Label done;
// Check for x % 0, sdiv might signal an exception. We have to deopt in this
// case because we can't return a NaN.
if (hmod->CheckFlag(HValue::kCanBeDivByZero)) {
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ cmp(right_reg, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kDivisionByZero);
}
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
// Check for kMinInt % -1, sdiv will return kMinInt, which is not what we
// want. We have to deopt if we care about -0, because we can't return that.
if (hmod->CheckFlag(HValue::kCanOverflow)) {
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
Label no_overflow_possible;
__ cmp(left_reg, Operand(kMinInt));
__ b(ne, &no_overflow_possible);
__ cmp(right_reg, Operand(-1));
if (hmod->CheckFlag(HValue::kBailoutOnMinusZero)) {
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
} else {
__ b(ne, &no_overflow_possible);
__ mov(result_reg, Operand::Zero());
__ jmp(&done);
}
__ bind(&no_overflow_possible);
}
// For 'r3 = r1 % r2' we can have the following ARM code:
// sdiv r3, r1, r2
// mls r3, r3, r2, r1
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ sdiv(result_reg, left_reg, right_reg);
__ Mls(result_reg, result_reg, right_reg, left_reg);
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
// If we care about -0, test if the dividend is <0 and the result is 0.
if (hmod->CheckFlag(HValue::kBailoutOnMinusZero)) {
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ cmp(result_reg, Operand::Zero());
__ b(ne, &done);
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ cmp(left_reg, Operand::Zero());
DeoptimizeIf(lt, instr, Deoptimizer::kMinusZero);
}
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ bind(&done);
} else {
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
// General case, without any SDIV support.
Register left_reg = ToRegister(instr->left());
Register right_reg = ToRegister(instr->right());
Register result_reg = ToRegister(instr->result());
Register scratch = scratch0();
DCHECK(!scratch.is(left_reg));
DCHECK(!scratch.is(right_reg));
DCHECK(!scratch.is(result_reg));
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
DwVfpRegister dividend = ToDoubleRegister(instr->temp());
DwVfpRegister divisor = ToDoubleRegister(instr->temp2());
DCHECK(!divisor.is(dividend));
LowDwVfpRegister quotient = double_scratch0();
DCHECK(!quotient.is(dividend));
DCHECK(!quotient.is(divisor));
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
Label done;
// Check for x % 0, we have to deopt in this case because we can't return a
// NaN.
if (hmod->CheckFlag(HValue::kCanBeDivByZero)) {
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ cmp(right_reg, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kDivisionByZero);
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
}
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ Move(result_reg, left_reg);
// Load the arguments in VFP registers. The divisor value is preloaded
// before. Be careful that 'right_reg' is only live on entry.
// TODO(svenpanne) The last comments seems to be wrong nowadays.
__ vmov(double_scratch0().low(), left_reg);
__ vcvt_f64_s32(dividend, double_scratch0().low());
__ vmov(double_scratch0().low(), right_reg);
__ vcvt_f64_s32(divisor, double_scratch0().low());
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
// We do not care about the sign of the divisor. Note that we still handle
// the kMinInt % -1 case correctly, though.
__ vabs(divisor, divisor);
// Compute the quotient and round it to a 32bit integer.
__ vdiv(quotient, dividend, divisor);
__ vcvt_s32_f64(quotient.low(), quotient);
__ vcvt_f64_s32(quotient, quotient.low());
// Compute the remainder in result.
__ vmul(double_scratch0(), divisor, quotient);
__ vcvt_s32_f64(double_scratch0().low(), double_scratch0());
__ vmov(scratch, double_scratch0().low());
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ sub(result_reg, left_reg, scratch, SetCC);
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
// If we care about -0, test if the dividend is <0 and the result is 0.
if (hmod->CheckFlag(HValue::kBailoutOnMinusZero)) {
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ b(ne, &done);
__ cmp(left_reg, Operand::Zero());
DeoptimizeIf(mi, instr, Deoptimizer::kMinusZero);
}
Improve code for integral modulus calculation. Depending on what we know about the right operand, we basically do 3 different things (and the code is actually structured this way): * If we statically know that the right operand is a power of 2, we do some bit fiddling instead of doing a "real" modulus calculation. This should actually be done on the Hydrogen level, not on the Lithium level, but this will be a separate CL. * If type feedback tells us that the right operand is a power of 2, we do the same as above, but guarded by conditional deoptimization to make sure that the assumption is still valid. In the long run, we should make this guard visible on the Hydrogen level to make it visible for GVN and other optimizations. * In the general case we only do the minimum steps necessary and don't try to be too clever, because cleverness actually slows us down on real-world code. If we look at the code gerators for LModI, we actually see that we basically have 3 (4 on ARM) fundamentally different translations. I don't really like lumping them together, they should probably be different Lithium instructions. For the time being, I restructured the generators to make this crystal-clear, at the cost of some duplication regarding the power-of-2 cases. This will go away when we do the strength reduction on the Hydrogen level, so I'd like to keep it as it is for now. Note that the MIPS part was only slightly restructured, there is still some work to do there. R=jkummerow@chromium.org Review URL: https://codereview.chromium.org/15769010 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15034 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-06-10 12:05:54 +00:00
__ bind(&done);
}
}
void LCodeGen::DoDivByPowerOf2I(LDivByPowerOf2I* instr) {
Register dividend = ToRegister(instr->dividend());
int32_t divisor = instr->divisor();
Register result = ToRegister(instr->result());
DCHECK(divisor == kMinInt || base::bits::IsPowerOfTwo32(Abs(divisor)));
DCHECK(!result.is(dividend));
// Check for (0 / -x) that will produce negative zero.
HDiv* hdiv = instr->hydrogen();
if (hdiv->CheckFlag(HValue::kBailoutOnMinusZero) && divisor < 0) {
__ cmp(dividend, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
}
// Check for (kMinInt / -1).
if (hdiv->CheckFlag(HValue::kCanOverflow) && divisor == -1) {
__ cmp(dividend, Operand(kMinInt));
DeoptimizeIf(eq, instr, Deoptimizer::kOverflow);
}
// Deoptimize if remainder will not be 0.
if (!hdiv->CheckFlag(HInstruction::kAllUsesTruncatingToInt32) &&
divisor != 1 && divisor != -1) {
int32_t mask = divisor < 0 ? -(divisor + 1) : (divisor - 1);
__ tst(dividend, Operand(mask));
DeoptimizeIf(ne, instr, Deoptimizer::kLostPrecision);
}
if (divisor == -1) { // Nice shortcut, not needed for correctness.
__ rsb(result, dividend, Operand(0));
return;
}
int32_t shift = WhichPowerOf2Abs(divisor);
if (shift == 0) {
__ mov(result, dividend);
} else if (shift == 1) {
__ add(result, dividend, Operand(dividend, LSR, 31));
} else {
__ mov(result, Operand(dividend, ASR, 31));
__ add(result, dividend, Operand(result, LSR, 32 - shift));
}
if (shift > 0) __ mov(result, Operand(result, ASR, shift));
if (divisor < 0) __ rsb(result, result, Operand(0));
}
void LCodeGen::DoDivByConstI(LDivByConstI* instr) {
Register dividend = ToRegister(instr->dividend());
int32_t divisor = instr->divisor();
Register result = ToRegister(instr->result());
DCHECK(!dividend.is(result));
if (divisor == 0) {
DeoptimizeIf(al, instr, Deoptimizer::kDivisionByZero);
return;
}
// Check for (0 / -x) that will produce negative zero.
HDiv* hdiv = instr->hydrogen();
if (hdiv->CheckFlag(HValue::kBailoutOnMinusZero) && divisor < 0) {
__ cmp(dividend, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
}
__ TruncatingDiv(result, dividend, Abs(divisor));
if (divisor < 0) __ rsb(result, result, Operand::Zero());
if (!hdiv->CheckFlag(HInstruction::kAllUsesTruncatingToInt32)) {
__ mov(ip, Operand(divisor));
__ smull(scratch0(), ip, result, ip);
__ sub(scratch0(), scratch0(), dividend, SetCC);
DeoptimizeIf(ne, instr, Deoptimizer::kLostPrecision);
}
}
// TODO(svenpanne) Refactor this to avoid code duplication with DoFlooringDivI.
void LCodeGen::DoDivI(LDivI* instr) {
HBinaryOperation* hdiv = instr->hydrogen();
Register dividend = ToRegister(instr->dividend());
Register divisor = ToRegister(instr->divisor());
Register result = ToRegister(instr->result());
// Check for x / 0.
if (hdiv->CheckFlag(HValue::kCanBeDivByZero)) {
__ cmp(divisor, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kDivisionByZero);
}
// Check for (0 / -x) that will produce negative zero.
if (hdiv->CheckFlag(HValue::kBailoutOnMinusZero)) {
Label positive;
if (!instr->hydrogen_value()->CheckFlag(HValue::kCanBeDivByZero)) {
// Do the test only if it hadn't be done above.
__ cmp(divisor, Operand::Zero());
}
__ b(pl, &positive);
__ cmp(dividend, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
__ bind(&positive);
}
// Check for (kMinInt / -1).
if (hdiv->CheckFlag(HValue::kCanOverflow) &&
(!CpuFeatures::IsSupported(SUDIV) ||
!hdiv->CheckFlag(HValue::kAllUsesTruncatingToInt32))) {
// We don't need to check for overflow when truncating with sdiv
// support because, on ARM, sdiv kMinInt, -1 -> kMinInt.
__ cmp(dividend, Operand(kMinInt));
__ cmp(divisor, Operand(-1), eq);
DeoptimizeIf(eq, instr, Deoptimizer::kOverflow);
}
if (CpuFeatures::IsSupported(SUDIV)) {
CpuFeatureScope scope(masm(), SUDIV);
__ sdiv(result, dividend, divisor);
} else {
DoubleRegister vleft = ToDoubleRegister(instr->temp());
DoubleRegister vright = double_scratch0();
__ vmov(double_scratch0().low(), dividend);
__ vcvt_f64_s32(vleft, double_scratch0().low());
__ vmov(double_scratch0().low(), divisor);
__ vcvt_f64_s32(vright, double_scratch0().low());
__ vdiv(vleft, vleft, vright); // vleft now contains the result.
__ vcvt_s32_f64(double_scratch0().low(), vleft);
__ vmov(result, double_scratch0().low());
}
if (!hdiv->CheckFlag(HValue::kAllUsesTruncatingToInt32)) {
// Compute remainder and deopt if it's not zero.
Register remainder = scratch0();
__ Mls(remainder, result, divisor, dividend);
__ cmp(remainder, Operand::Zero());
DeoptimizeIf(ne, instr, Deoptimizer::kLostPrecision);
}
}
void LCodeGen::DoMultiplyAddD(LMultiplyAddD* instr) {
DwVfpRegister addend = ToDoubleRegister(instr->addend());
DwVfpRegister multiplier = ToDoubleRegister(instr->multiplier());
DwVfpRegister multiplicand = ToDoubleRegister(instr->multiplicand());
// This is computed in-place.
DCHECK(addend.is(ToDoubleRegister(instr->result())));
__ vmla(addend, multiplier, multiplicand);
}
void LCodeGen::DoMultiplySubD(LMultiplySubD* instr) {
DwVfpRegister minuend = ToDoubleRegister(instr->minuend());
DwVfpRegister multiplier = ToDoubleRegister(instr->multiplier());
DwVfpRegister multiplicand = ToDoubleRegister(instr->multiplicand());
// This is computed in-place.
DCHECK(minuend.is(ToDoubleRegister(instr->result())));
__ vmls(minuend, multiplier, multiplicand);
}
void LCodeGen::DoFlooringDivByPowerOf2I(LFlooringDivByPowerOf2I* instr) {
Register dividend = ToRegister(instr->dividend());
Register result = ToRegister(instr->result());
int32_t divisor = instr->divisor();
// If the divisor is 1, return the dividend.
if (divisor == 1) {
__ Move(result, dividend);
return;
}
// If the divisor is positive, things are easy: There can be no deopts and we
// can simply do an arithmetic right shift.
int32_t shift = WhichPowerOf2Abs(divisor);
if (divisor > 1) {
__ mov(result, Operand(dividend, ASR, shift));
return;
}
// If the divisor is negative, we have to negate and handle edge cases.
__ rsb(result, dividend, Operand::Zero(), SetCC);
if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
}
// Dividing by -1 is basically negation, unless we overflow.
if (divisor == -1) {
if (instr->hydrogen()->CheckFlag(HValue::kLeftCanBeMinInt)) {
DeoptimizeIf(vs, instr, Deoptimizer::kOverflow);
}
return;
}
// If the negation could not overflow, simply shifting is OK.
if (!instr->hydrogen()->CheckFlag(HValue::kLeftCanBeMinInt)) {
__ mov(result, Operand(result, ASR, shift));
return;
}
__ mov(result, Operand(kMinInt / divisor), LeaveCC, vs);
__ mov(result, Operand(result, ASR, shift), LeaveCC, vc);
}
void LCodeGen::DoFlooringDivByConstI(LFlooringDivByConstI* instr) {
Register dividend = ToRegister(instr->dividend());
int32_t divisor = instr->divisor();
Register result = ToRegister(instr->result());
DCHECK(!dividend.is(result));
if (divisor == 0) {
DeoptimizeIf(al, instr, Deoptimizer::kDivisionByZero);
return;
}
// Check for (0 / -x) that will produce negative zero.
HMathFloorOfDiv* hdiv = instr->hydrogen();
if (hdiv->CheckFlag(HValue::kBailoutOnMinusZero) && divisor < 0) {
__ cmp(dividend, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
}
// Easy case: We need no dynamic check for the dividend and the flooring
// division is the same as the truncating division.
if ((divisor > 0 && !hdiv->CheckFlag(HValue::kLeftCanBeNegative)) ||
(divisor < 0 && !hdiv->CheckFlag(HValue::kLeftCanBePositive))) {
__ TruncatingDiv(result, dividend, Abs(divisor));
if (divisor < 0) __ rsb(result, result, Operand::Zero());
return;
}
// In the general case we may need to adjust before and after the truncating
// division to get a flooring division.
Register temp = ToRegister(instr->temp());
DCHECK(!temp.is(dividend) && !temp.is(result));
Label needs_adjustment, done;
__ cmp(dividend, Operand::Zero());
__ b(divisor > 0 ? lt : gt, &needs_adjustment);
__ TruncatingDiv(result, dividend, Abs(divisor));
if (divisor < 0) __ rsb(result, result, Operand::Zero());
__ jmp(&done);
__ bind(&needs_adjustment);
__ add(temp, dividend, Operand(divisor > 0 ? 1 : -1));
__ TruncatingDiv(result, temp, Abs(divisor));
if (divisor < 0) __ rsb(result, result, Operand::Zero());
__ sub(result, result, Operand(1));
__ bind(&done);
}
// TODO(svenpanne) Refactor this to avoid code duplication with DoDivI.
void LCodeGen::DoFlooringDivI(LFlooringDivI* instr) {
HBinaryOperation* hdiv = instr->hydrogen();
Register left = ToRegister(instr->dividend());
Register right = ToRegister(instr->divisor());
Register result = ToRegister(instr->result());
// Check for x / 0.
if (hdiv->CheckFlag(HValue::kCanBeDivByZero)) {
__ cmp(right, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kDivisionByZero);
}
// Check for (0 / -x) that will produce negative zero.
if (hdiv->CheckFlag(HValue::kBailoutOnMinusZero)) {
Label positive;
if (!instr->hydrogen_value()->CheckFlag(HValue::kCanBeDivByZero)) {
// Do the test only if it hadn't be done above.
__ cmp(right, Operand::Zero());
}
__ b(pl, &positive);
__ cmp(left, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
__ bind(&positive);
}
// Check for (kMinInt / -1).
if (hdiv->CheckFlag(HValue::kCanOverflow) &&
(!CpuFeatures::IsSupported(SUDIV) ||
!hdiv->CheckFlag(HValue::kAllUsesTruncatingToInt32))) {
// We don't need to check for overflow when truncating with sdiv
// support because, on ARM, sdiv kMinInt, -1 -> kMinInt.
__ cmp(left, Operand(kMinInt));
__ cmp(right, Operand(-1), eq);
DeoptimizeIf(eq, instr, Deoptimizer::kOverflow);
}
if (CpuFeatures::IsSupported(SUDIV)) {
CpuFeatureScope scope(masm(), SUDIV);
__ sdiv(result, left, right);
} else {
DoubleRegister vleft = ToDoubleRegister(instr->temp());
DoubleRegister vright = double_scratch0();
__ vmov(double_scratch0().low(), left);
__ vcvt_f64_s32(vleft, double_scratch0().low());
__ vmov(double_scratch0().low(), right);
__ vcvt_f64_s32(vright, double_scratch0().low());
__ vdiv(vleft, vleft, vright); // vleft now contains the result.
__ vcvt_s32_f64(double_scratch0().low(), vleft);
__ vmov(result, double_scratch0().low());
}
Label done;
Register remainder = scratch0();
__ Mls(remainder, result, right, left);
__ cmp(remainder, Operand::Zero());
__ b(eq, &done);
__ eor(remainder, remainder, Operand(right));
__ add(result, result, Operand(remainder, ASR, 31));
__ bind(&done);
}
void LCodeGen::DoMulI(LMulI* instr) {
Register result = ToRegister(instr->result());
// Note that result may alias left.
Register left = ToRegister(instr->left());
LOperand* right_op = instr->right();
bool bailout_on_minus_zero =
instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero);
bool overflow = instr->hydrogen()->CheckFlag(HValue::kCanOverflow);
if (right_op->IsConstantOperand()) {
int32_t constant = ToInteger32(LConstantOperand::cast(right_op));
if (bailout_on_minus_zero && (constant < 0)) {
// The case of a null constant will be handled separately.
// If constant is negative and left is null, the result should be -0.
__ cmp(left, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
}
switch (constant) {
case -1:
if (overflow) {
__ rsb(result, left, Operand::Zero(), SetCC);
DeoptimizeIf(vs, instr, Deoptimizer::kOverflow);
} else {
__ rsb(result, left, Operand::Zero());
}
break;
case 0:
if (bailout_on_minus_zero) {
// If left is strictly negative and the constant is null, the
// result is -0. Deoptimize if required, otherwise return 0.
__ cmp(left, Operand::Zero());
DeoptimizeIf(mi, instr, Deoptimizer::kMinusZero);
}
__ mov(result, Operand::Zero());
break;
case 1:
__ Move(result, left);
break;
default:
// Multiplying by powers of two and powers of two plus or minus
// one can be done faster with shifted operands.
// For other constants we emit standard code.
int32_t mask = constant >> 31;
uint32_t constant_abs = (constant + mask) ^ mask;
if (base::bits::IsPowerOfTwo32(constant_abs)) {
int32_t shift = WhichPowerOf2(constant_abs);
__ mov(result, Operand(left, LSL, shift));
// Correct the sign of the result is the constant is negative.
if (constant < 0) __ rsb(result, result, Operand::Zero());
} else if (base::bits::IsPowerOfTwo32(constant_abs - 1)) {
int32_t shift = WhichPowerOf2(constant_abs - 1);
__ add(result, left, Operand(left, LSL, shift));
// Correct the sign of the result is the constant is negative.
if (constant < 0) __ rsb(result, result, Operand::Zero());
} else if (base::bits::IsPowerOfTwo32(constant_abs + 1)) {
int32_t shift = WhichPowerOf2(constant_abs + 1);
__ rsb(result, left, Operand(left, LSL, shift));
// Correct the sign of the result is the constant is negative.
if (constant < 0) __ rsb(result, result, Operand::Zero());
} else {
// Generate standard code.
__ mov(ip, Operand(constant));
__ mul(result, left, ip);
}
}
} else {
DCHECK(right_op->IsRegister());
Register right = ToRegister(right_op);
if (overflow) {
Register scratch = scratch0();
// scratch:result = left * right.
if (instr->hydrogen()->representation().IsSmi()) {
__ SmiUntag(result, left);
__ smull(result, scratch, result, right);
} else {
__ smull(result, scratch, left, right);
}
__ cmp(scratch, Operand(result, ASR, 31));
DeoptimizeIf(ne, instr, Deoptimizer::kOverflow);
} else {
if (instr->hydrogen()->representation().IsSmi()) {
__ SmiUntag(result, left);
__ mul(result, result, right);
} else {
__ mul(result, left, right);
}
}
if (bailout_on_minus_zero) {
Label done;
__ teq(left, Operand(right));
__ b(pl, &done);
// Bail out if the result is minus zero.
__ cmp(result, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
__ bind(&done);
}
}
}
void LCodeGen::DoBitI(LBitI* instr) {
LOperand* left_op = instr->left();
LOperand* right_op = instr->right();
DCHECK(left_op->IsRegister());
Register left = ToRegister(left_op);
Register result = ToRegister(instr->result());
Operand right(no_reg);
if (right_op->IsStackSlot()) {
right = Operand(EmitLoadRegister(right_op, ip));
} else {
DCHECK(right_op->IsRegister() || right_op->IsConstantOperand());
right = ToOperand(right_op);
}
switch (instr->op()) {
case Token::BIT_AND:
__ and_(result, left, right);
break;
case Token::BIT_OR:
__ orr(result, left, right);
break;
case Token::BIT_XOR:
if (right_op->IsConstantOperand() && right.immediate() == int32_t(~0)) {
__ mvn(result, Operand(left));
} else {
__ eor(result, left, right);
}
break;
default:
UNREACHABLE();
break;
}
}
void LCodeGen::DoShiftI(LShiftI* instr) {
// Both 'left' and 'right' are "used at start" (see LCodeGen::DoShift), so
// result may alias either of them.
LOperand* right_op = instr->right();
Register left = ToRegister(instr->left());
Register result = ToRegister(instr->result());
Register scratch = scratch0();
if (right_op->IsRegister()) {
// Mask the right_op operand.
__ and_(scratch, ToRegister(right_op), Operand(0x1F));
switch (instr->op()) {
case Token::ROR:
__ mov(result, Operand(left, ROR, scratch));
break;
case Token::SAR:
__ mov(result, Operand(left, ASR, scratch));
break;
case Token::SHR:
if (instr->can_deopt()) {
__ mov(result, Operand(left, LSR, scratch), SetCC);
DeoptimizeIf(mi, instr, Deoptimizer::kNegativeValue);
} else {
__ mov(result, Operand(left, LSR, scratch));
}
break;
case Token::SHL:
__ mov(result, Operand(left, LSL, scratch));
break;
default:
UNREACHABLE();
break;
}
} else {
// Mask the right_op operand.
int value = ToInteger32(LConstantOperand::cast(right_op));
uint8_t shift_count = static_cast<uint8_t>(value & 0x1F);
switch (instr->op()) {
case Token::ROR:
if (shift_count != 0) {
__ mov(result, Operand(left, ROR, shift_count));
} else {
__ Move(result, left);
}
break;
case Token::SAR:
if (shift_count != 0) {
__ mov(result, Operand(left, ASR, shift_count));
} else {
__ Move(result, left);
}
break;
case Token::SHR:
if (shift_count != 0) {
__ mov(result, Operand(left, LSR, shift_count));
} else {
if (instr->can_deopt()) {
__ tst(left, Operand(0x80000000));
DeoptimizeIf(ne, instr, Deoptimizer::kNegativeValue);
}
__ Move(result, left);
}
break;
case Token::SHL:
if (shift_count != 0) {
if (instr->hydrogen_value()->representation().IsSmi() &&
instr->can_deopt()) {
if (shift_count != 1) {
__ mov(result, Operand(left, LSL, shift_count - 1));
__ SmiTag(result, result, SetCC);
} else {
__ SmiTag(result, left, SetCC);
}
DeoptimizeIf(vs, instr, Deoptimizer::kOverflow);
} else {
__ mov(result, Operand(left, LSL, shift_count));
}
} else {
__ Move(result, left);
}
break;
default:
UNREACHABLE();
break;
}
}
}
void LCodeGen::DoSubI(LSubI* instr) {
LOperand* left = instr->left();
LOperand* right = instr->right();
LOperand* result = instr->result();
bool can_overflow = instr->hydrogen()->CheckFlag(HValue::kCanOverflow);
SBit set_cond = can_overflow ? SetCC : LeaveCC;
if (right->IsStackSlot()) {
Register right_reg = EmitLoadRegister(right, ip);
__ sub(ToRegister(result), ToRegister(left), Operand(right_reg), set_cond);
} else {
DCHECK(right->IsRegister() || right->IsConstantOperand());
__ sub(ToRegister(result), ToRegister(left), ToOperand(right), set_cond);
}
if (can_overflow) {
DeoptimizeIf(vs, instr, Deoptimizer::kOverflow);
}
}
void LCodeGen::DoRSubI(LRSubI* instr) {
LOperand* left = instr->left();
LOperand* right = instr->right();
LOperand* result = instr->result();
bool can_overflow = instr->hydrogen()->CheckFlag(HValue::kCanOverflow);
SBit set_cond = can_overflow ? SetCC : LeaveCC;
if (right->IsStackSlot()) {
Register right_reg = EmitLoadRegister(right, ip);
__ rsb(ToRegister(result), ToRegister(left), Operand(right_reg), set_cond);
} else {
DCHECK(right->IsRegister() || right->IsConstantOperand());
__ rsb(ToRegister(result), ToRegister(left), ToOperand(right), set_cond);
}
if (can_overflow) {
DeoptimizeIf(vs, instr, Deoptimizer::kOverflow);
}
}
void LCodeGen::DoConstantI(LConstantI* instr) {
__ mov(ToRegister(instr->result()), Operand(instr->value()));
}
void LCodeGen::DoConstantS(LConstantS* instr) {
__ mov(ToRegister(instr->result()), Operand(instr->value()));
}
void LCodeGen::DoConstantD(LConstantD* instr) {
DCHECK(instr->result()->IsDoubleRegister());
DwVfpRegister result = ToDoubleRegister(instr->result());
#if V8_HOST_ARCH_IA32
// Need some crappy work-around for x87 sNaN -> qNaN breakage in simulator
// builds.
uint64_t bits = instr->bits();
if ((bits & V8_UINT64_C(0x7FF8000000000000)) ==
V8_UINT64_C(0x7FF0000000000000)) {
uint32_t lo = static_cast<uint32_t>(bits);
uint32_t hi = static_cast<uint32_t>(bits >> 32);
__ mov(ip, Operand(lo));
__ mov(scratch0(), Operand(hi));
__ vmov(result, ip, scratch0());
return;
}
#endif
double v = instr->value();
__ Vmov(result, v, scratch0());
}
void LCodeGen::DoConstantE(LConstantE* instr) {
__ mov(ToRegister(instr->result()), Operand(instr->value()));
}
void LCodeGen::DoConstantT(LConstantT* instr) {
Handle<Object> object = instr->value(isolate());
AllowDeferredHandleDereference smi_check;
__ Move(ToRegister(instr->result()), object);
}
void LCodeGen::DoMapEnumLength(LMapEnumLength* instr) {
Register result = ToRegister(instr->result());
Register map = ToRegister(instr->value());
__ EnumLength(result, map);
}
void LCodeGen::DoDateField(LDateField* instr) {
Register object = ToRegister(instr->date());
Register result = ToRegister(instr->result());
Register scratch = ToRegister(instr->temp());
Smi* index = instr->index();
Label runtime, done;
DCHECK(object.is(result));
DCHECK(object.is(r0));
DCHECK(!scratch.is(scratch0()));
DCHECK(!scratch.is(object));
__ SmiTst(object);
DeoptimizeIf(eq, instr, Deoptimizer::kSmi);
__ CompareObjectType(object, scratch, scratch, JS_DATE_TYPE);
DeoptimizeIf(ne, instr, Deoptimizer::kNotADateObject);
if (index->value() == 0) {
__ ldr(result, FieldMemOperand(object, JSDate::kValueOffset));
} else {
if (index->value() < JSDate::kFirstUncachedField) {
ExternalReference stamp = ExternalReference::date_cache_stamp(isolate());
__ mov(scratch, Operand(stamp));
__ ldr(scratch, MemOperand(scratch));
__ ldr(scratch0(), FieldMemOperand(object, JSDate::kCacheStampOffset));
__ cmp(scratch, scratch0());
__ b(ne, &runtime);
__ ldr(result, FieldMemOperand(object, JSDate::kValueOffset +
kPointerSize * index->value()));
__ jmp(&done);
}
__ bind(&runtime);
__ PrepareCallCFunction(2, scratch);
__ mov(r1, Operand(index));
__ CallCFunction(ExternalReference::get_date_field_function(isolate()), 2);
__ bind(&done);
}
}
MemOperand LCodeGen::BuildSeqStringOperand(Register string,
LOperand* index,
String::Encoding encoding) {
if (index->IsConstantOperand()) {
int offset = ToInteger32(LConstantOperand::cast(index));
if (encoding == String::TWO_BYTE_ENCODING) {
offset *= kUC16Size;
}
STATIC_ASSERT(kCharSize == 1);
return FieldMemOperand(string, SeqString::kHeaderSize + offset);
}
Register scratch = scratch0();
DCHECK(!scratch.is(string));
DCHECK(!scratch.is(ToRegister(index)));
if (encoding == String::ONE_BYTE_ENCODING) {
__ add(scratch, string, Operand(ToRegister(index)));
} else {
STATIC_ASSERT(kUC16Size == 2);
__ add(scratch, string, Operand(ToRegister(index), LSL, 1));
}
return FieldMemOperand(scratch, SeqString::kHeaderSize);
}
void LCodeGen::DoSeqStringGetChar(LSeqStringGetChar* instr) {
String::Encoding encoding = instr->hydrogen()->encoding();
Register string = ToRegister(instr->string());
Register result = ToRegister(instr->result());
if (FLAG_debug_code) {
Register scratch = scratch0();
__ ldr(scratch, FieldMemOperand(string, HeapObject::kMapOffset));
__ ldrb(scratch, FieldMemOperand(scratch, Map::kInstanceTypeOffset));
__ and_(scratch, scratch,
Operand(kStringRepresentationMask | kStringEncodingMask));
static const uint32_t one_byte_seq_type = kSeqStringTag | kOneByteStringTag;
static const uint32_t two_byte_seq_type = kSeqStringTag | kTwoByteStringTag;
__ cmp(scratch, Operand(encoding == String::ONE_BYTE_ENCODING
? one_byte_seq_type : two_byte_seq_type));
__ Check(eq, kUnexpectedStringType);
}
MemOperand operand = BuildSeqStringOperand(string, instr->index(), encoding);
if (encoding == String::ONE_BYTE_ENCODING) {
__ ldrb(result, operand);
} else {
__ ldrh(result, operand);
}
}
void LCodeGen::DoSeqStringSetChar(LSeqStringSetChar* instr) {
String::Encoding encoding = instr->hydrogen()->encoding();
Register string = ToRegister(instr->string());
Register value = ToRegister(instr->value());
if (FLAG_debug_code) {
Register index = ToRegister(instr->index());
static const uint32_t one_byte_seq_type = kSeqStringTag | kOneByteStringTag;
static const uint32_t two_byte_seq_type = kSeqStringTag | kTwoByteStringTag;
int encoding_mask =
instr->hydrogen()->encoding() == String::ONE_BYTE_ENCODING
? one_byte_seq_type : two_byte_seq_type;
__ EmitSeqStringSetCharCheck(string, index, value, encoding_mask);
}
MemOperand operand = BuildSeqStringOperand(string, instr->index(), encoding);
if (encoding == String::ONE_BYTE_ENCODING) {
__ strb(value, operand);
} else {
__ strh(value, operand);
}
}
void LCodeGen::DoAddI(LAddI* instr) {
LOperand* left = instr->left();
LOperand* right = instr->right();
LOperand* result = instr->result();
bool can_overflow = instr->hydrogen()->CheckFlag(HValue::kCanOverflow);
SBit set_cond = can_overflow ? SetCC : LeaveCC;
if (right->IsStackSlot()) {
Register right_reg = EmitLoadRegister(right, ip);
__ add(ToRegister(result), ToRegister(left), Operand(right_reg), set_cond);
} else {
DCHECK(right->IsRegister() || right->IsConstantOperand());
__ add(ToRegister(result), ToRegister(left), ToOperand(right), set_cond);
}
if (can_overflow) {
DeoptimizeIf(vs, instr, Deoptimizer::kOverflow);
}
}
void LCodeGen::DoMathMinMax(LMathMinMax* instr) {
LOperand* left = instr->left();
LOperand* right = instr->right();
HMathMinMax::Operation operation = instr->hydrogen()->operation();
if (instr->hydrogen()->representation().IsSmiOrInteger32()) {
Condition condition = (operation == HMathMinMax::kMathMin) ? le : ge;
Register left_reg = ToRegister(left);
Operand right_op = (right->IsRegister() || right->IsConstantOperand())
? ToOperand(right)
: Operand(EmitLoadRegister(right, ip));
Register result_reg = ToRegister(instr->result());
__ cmp(left_reg, right_op);
__ Move(result_reg, left_reg, condition);
__ mov(result_reg, right_op, LeaveCC, NegateCondition(condition));
} else {
DCHECK(instr->hydrogen()->representation().IsDouble());
DwVfpRegister left_reg = ToDoubleRegister(left);
DwVfpRegister right_reg = ToDoubleRegister(right);
DwVfpRegister result_reg = ToDoubleRegister(instr->result());
Label result_is_nan, return_left, return_right, check_zero, done;
__ VFPCompareAndSetFlags(left_reg, right_reg);
if (operation == HMathMinMax::kMathMin) {
__ b(mi, &return_left);
__ b(gt, &return_right);
} else {
__ b(mi, &return_right);
__ b(gt, &return_left);
}
__ b(vs, &result_is_nan);
// Left equals right => check for -0.
__ VFPCompareAndSetFlags(left_reg, 0.0);
if (left_reg.is(result_reg) || right_reg.is(result_reg)) {
__ b(ne, &done); // left == right != 0.
} else {
__ b(ne, &return_left); // left == right != 0.
}
// At this point, both left and right are either 0 or -0.
if (operation == HMathMinMax::kMathMin) {
// We could use a single 'vorr' instruction here if we had NEON support.
__ vneg(left_reg, left_reg);
__ vsub(result_reg, left_reg, right_reg);
__ vneg(result_reg, result_reg);
} else {
// Since we operate on +0 and/or -0, vadd and vand have the same effect;
// the decision for vadd is easy because vand is a NEON instruction.
__ vadd(result_reg, left_reg, right_reg);
}
__ b(&done);
__ bind(&result_is_nan);
__ vadd(result_reg, left_reg, right_reg);
__ b(&done);
__ bind(&return_right);
__ Move(result_reg, right_reg);
if (!left_reg.is(result_reg)) {
__ b(&done);
}
__ bind(&return_left);
__ Move(result_reg, left_reg);
__ bind(&done);
}
}
void LCodeGen::DoArithmeticD(LArithmeticD* instr) {
DwVfpRegister left = ToDoubleRegister(instr->left());
DwVfpRegister right = ToDoubleRegister(instr->right());
DwVfpRegister result = ToDoubleRegister(instr->result());
switch (instr->op()) {
case Token::ADD:
__ vadd(result, left, right);
break;
case Token::SUB:
__ vsub(result, left, right);
break;
case Token::MUL:
__ vmul(result, left, right);
break;
case Token::DIV:
__ vdiv(result, left, right);
break;
case Token::MOD: {
__ PrepareCallCFunction(0, 2, scratch0());
__ MovToFloatParameters(left, right);
__ CallCFunction(
ExternalReference::mod_two_doubles_operation(isolate()),
0, 2);
// Move the result in the double result register.
__ MovFromFloatResult(result);
break;
}
default:
UNREACHABLE();
break;
}
}
void LCodeGen::DoArithmeticT(LArithmeticT* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->left()).is(r1));
DCHECK(ToRegister(instr->right()).is(r0));
DCHECK(ToRegister(instr->result()).is(r0));
Handle<Code> code = CodeFactory::BinaryOpIC(isolate(), instr->op()).code();
// Block literal pool emission to ensure nop indicating no inlined smi code
// is in the correct position.
Assembler::BlockConstPoolScope block_const_pool(masm());
CallCode(code, RelocInfo::CODE_TARGET, instr);
}
template<class InstrType>
void LCodeGen::EmitBranch(InstrType instr, Condition condition) {
int left_block = instr->TrueDestination(chunk_);
int right_block = instr->FalseDestination(chunk_);
int next_block = GetNextEmittedBlock();
if (right_block == left_block || condition == al) {
EmitGoto(left_block);
} else if (left_block == next_block) {
__ b(NegateCondition(condition), chunk_->GetAssemblyLabel(right_block));
} else if (right_block == next_block) {
__ b(condition, chunk_->GetAssemblyLabel(left_block));
} else {
__ b(condition, chunk_->GetAssemblyLabel(left_block));
__ b(chunk_->GetAssemblyLabel(right_block));
}
}
template<class InstrType>
void LCodeGen::EmitFalseBranch(InstrType instr, Condition condition) {
int false_block = instr->FalseDestination(chunk_);
__ b(condition, chunk_->GetAssemblyLabel(false_block));
}
void LCodeGen::DoDebugBreak(LDebugBreak* instr) {
__ stop("LBreak");
}
void LCodeGen::DoBranch(LBranch* instr) {
Representation r = instr->hydrogen()->value()->representation();
if (r.IsInteger32() || r.IsSmi()) {
DCHECK(!info()->IsStub());
Register reg = ToRegister(instr->value());
__ cmp(reg, Operand::Zero());
EmitBranch(instr, ne);
} else if (r.IsDouble()) {
DCHECK(!info()->IsStub());
DwVfpRegister reg = ToDoubleRegister(instr->value());
// Test the double value. Zero and NaN are false.
__ VFPCompareAndSetFlags(reg, 0.0);
__ cmp(r0, r0, vs); // If NaN, set the Z flag. (NaN -> false)
EmitBranch(instr, ne);
} else {
DCHECK(r.IsTagged());
Register reg = ToRegister(instr->value());
HType type = instr->hydrogen()->value()->type();
if (type.IsBoolean()) {
DCHECK(!info()->IsStub());
__ CompareRoot(reg, Heap::kTrueValueRootIndex);
EmitBranch(instr, eq);
} else if (type.IsSmi()) {
DCHECK(!info()->IsStub());
__ cmp(reg, Operand::Zero());
EmitBranch(instr, ne);
} else if (type.IsJSArray()) {
DCHECK(!info()->IsStub());
EmitBranch(instr, al);
} else if (type.IsHeapNumber()) {
DCHECK(!info()->IsStub());
DwVfpRegister dbl_scratch = double_scratch0();
__ vldr(dbl_scratch, FieldMemOperand(reg, HeapNumber::kValueOffset));
// Test the double value. Zero and NaN are false.
__ VFPCompareAndSetFlags(dbl_scratch, 0.0);
__ cmp(r0, r0, vs); // If NaN, set the Z flag. (NaN)
EmitBranch(instr, ne);
} else if (type.IsString()) {
DCHECK(!info()->IsStub());
__ ldr(ip, FieldMemOperand(reg, String::kLengthOffset));
__ cmp(ip, Operand::Zero());
EmitBranch(instr, ne);
} else {
ToBooleanStub::Types expected = instr->hydrogen()->expected_input_types();
// Avoid deopts in the case where we've never executed this path before.
if (expected.IsEmpty()) expected = ToBooleanStub::Types::Generic();
if (expected.Contains(ToBooleanStub::UNDEFINED)) {
// undefined -> false.
__ CompareRoot(reg, Heap::kUndefinedValueRootIndex);
__ b(eq, instr->FalseLabel(chunk_));
}
if (expected.Contains(ToBooleanStub::BOOLEAN)) {
// Boolean -> its value.
__ CompareRoot(reg, Heap::kTrueValueRootIndex);
__ b(eq, instr->TrueLabel(chunk_));
__ CompareRoot(reg, Heap::kFalseValueRootIndex);
__ b(eq, instr->FalseLabel(chunk_));
}
if (expected.Contains(ToBooleanStub::NULL_TYPE)) {
// 'null' -> false.
__ CompareRoot(reg, Heap::kNullValueRootIndex);
__ b(eq, instr->FalseLabel(chunk_));
}
if (expected.Contains(ToBooleanStub::SMI)) {
// Smis: 0 -> false, all other -> true.
__ cmp(reg, Operand::Zero());
__ b(eq, instr->FalseLabel(chunk_));
__ JumpIfSmi(reg, instr->TrueLabel(chunk_));
} else if (expected.NeedsMap()) {
// If we need a map later and have a Smi -> deopt.
__ SmiTst(reg);
DeoptimizeIf(eq, instr, Deoptimizer::kSmi);
}
const Register map = scratch0();
if (expected.NeedsMap()) {
__ ldr(map, FieldMemOperand(reg, HeapObject::kMapOffset));
if (expected.CanBeUndetectable()) {
// Undetectable -> false.
__ ldrb(ip, FieldMemOperand(map, Map::kBitFieldOffset));
__ tst(ip, Operand(1 << Map::kIsUndetectable));
__ b(ne, instr->FalseLabel(chunk_));
}
}
if (expected.Contains(ToBooleanStub::SPEC_OBJECT)) {
// spec object -> true.
__ CompareInstanceType(map, ip, FIRST_SPEC_OBJECT_TYPE);
__ b(ge, instr->TrueLabel(chunk_));
}
if (expected.Contains(ToBooleanStub::STRING)) {
// String value -> false iff empty.
Label not_string;
__ CompareInstanceType(map, ip, FIRST_NONSTRING_TYPE);
__ b(ge, &not_string);
__ ldr(ip, FieldMemOperand(reg, String::kLengthOffset));
__ cmp(ip, Operand::Zero());
__ b(ne, instr->TrueLabel(chunk_));
__ b(instr->FalseLabel(chunk_));
__ bind(&not_string);
}
if (expected.Contains(ToBooleanStub::SYMBOL)) {
// Symbol value -> true.
__ CompareInstanceType(map, ip, SYMBOL_TYPE);
__ b(eq, instr->TrueLabel(chunk_));
}
if (expected.Contains(ToBooleanStub::HEAP_NUMBER)) {
// heap number -> false iff +0, -0, or NaN.
DwVfpRegister dbl_scratch = double_scratch0();
Label not_heap_number;
__ CompareRoot(map, Heap::kHeapNumberMapRootIndex);
__ b(ne, &not_heap_number);
__ vldr(dbl_scratch, FieldMemOperand(reg, HeapNumber::kValueOffset));
__ VFPCompareAndSetFlags(dbl_scratch, 0.0);
__ cmp(r0, r0, vs); // NaN -> false.
__ b(eq, instr->FalseLabel(chunk_)); // +0, -0 -> false.
__ b(instr->TrueLabel(chunk_));
__ bind(&not_heap_number);
}
if (!expected.IsGeneric()) {
// We've seen something for the first time -> deopt.
// This can only happen if we are not generic already.
DeoptimizeIf(al, instr, Deoptimizer::kUnexpectedObject);
}
}
}
}
void LCodeGen::EmitGoto(int block) {
if (!IsNextEmittedBlock(block)) {
__ jmp(chunk_->GetAssemblyLabel(LookupDestination(block)));
}
}
void LCodeGen::DoGoto(LGoto* instr) {
EmitGoto(instr->block_id());
}
Condition LCodeGen::TokenToCondition(Token::Value op, bool is_unsigned) {
Condition cond = kNoCondition;
switch (op) {
case Token::EQ:
case Token::EQ_STRICT:
cond = eq;
break;
case Token::NE:
case Token::NE_STRICT:
cond = ne;
break;
case Token::LT:
cond = is_unsigned ? lo : lt;
break;
case Token::GT:
cond = is_unsigned ? hi : gt;
break;
case Token::LTE:
cond = is_unsigned ? ls : le;
break;
case Token::GTE:
cond = is_unsigned ? hs : ge;
break;
case Token::IN:
case Token::INSTANCEOF:
default:
UNREACHABLE();
}
return cond;
}
void LCodeGen::DoCompareNumericAndBranch(LCompareNumericAndBranch* instr) {
LOperand* left = instr->left();
LOperand* right = instr->right();
bool is_unsigned =
instr->hydrogen()->left()->CheckFlag(HInstruction::kUint32) ||
instr->hydrogen()->right()->CheckFlag(HInstruction::kUint32);
Condition cond = TokenToCondition(instr->op(), is_unsigned);
if (left->IsConstantOperand() && right->IsConstantOperand()) {
// We can statically evaluate the comparison.
double left_val = ToDouble(LConstantOperand::cast(left));
double right_val = ToDouble(LConstantOperand::cast(right));
int next_block = EvalComparison(instr->op(), left_val, right_val) ?
instr->TrueDestination(chunk_) : instr->FalseDestination(chunk_);
EmitGoto(next_block);
} else {
if (instr->is_double()) {
// Compare left and right operands as doubles and load the
// resulting flags into the normal status register.
__ VFPCompareAndSetFlags(ToDoubleRegister(left), ToDoubleRegister(right));
// If a NaN is involved, i.e. the result is unordered (V set),
// jump to false block label.
__ b(vs, instr->FalseLabel(chunk_));
} else {
if (right->IsConstantOperand()) {
int32_t value = ToInteger32(LConstantOperand::cast(right));
if (instr->hydrogen_value()->representation().IsSmi()) {
__ cmp(ToRegister(left), Operand(Smi::FromInt(value)));
} else {
__ cmp(ToRegister(left), Operand(value));
}
} else if (left->IsConstantOperand()) {
int32_t value = ToInteger32(LConstantOperand::cast(left));
if (instr->hydrogen_value()->representation().IsSmi()) {
__ cmp(ToRegister(right), Operand(Smi::FromInt(value)));
} else {
__ cmp(ToRegister(right), Operand(value));
}
// We commuted the operands, so commute the condition.
cond = CommuteCondition(cond);
} else {
__ cmp(ToRegister(left), ToRegister(right));
}
}
EmitBranch(instr, cond);
}
}
void LCodeGen::DoCmpObjectEqAndBranch(LCmpObjectEqAndBranch* instr) {
Register left = ToRegister(instr->left());
Register right = ToRegister(instr->right());
__ cmp(left, Operand(right));
EmitBranch(instr, eq);
}
void LCodeGen::DoCmpHoleAndBranch(LCmpHoleAndBranch* instr) {
if (instr->hydrogen()->representation().IsTagged()) {
Register input_reg = ToRegister(instr->object());
__ mov(ip, Operand(factory()->the_hole_value()));
__ cmp(input_reg, ip);
EmitBranch(instr, eq);
return;
}
DwVfpRegister input_reg = ToDoubleRegister(instr->object());
__ VFPCompareAndSetFlags(input_reg, input_reg);
EmitFalseBranch(instr, vc);
Register scratch = scratch0();
__ VmovHigh(scratch, input_reg);
__ cmp(scratch, Operand(kHoleNanUpper32));
EmitBranch(instr, eq);
}
void LCodeGen::DoCompareMinusZeroAndBranch(LCompareMinusZeroAndBranch* instr) {
Representation rep = instr->hydrogen()->value()->representation();
DCHECK(!rep.IsInteger32());
Register scratch = ToRegister(instr->temp());
if (rep.IsDouble()) {
DwVfpRegister value = ToDoubleRegister(instr->value());
__ VFPCompareAndSetFlags(value, 0.0);
EmitFalseBranch(instr, ne);
__ VmovHigh(scratch, value);
__ cmp(scratch, Operand(0x80000000));
} else {
Register value = ToRegister(instr->value());
__ CheckMap(value,
scratch,
Heap::kHeapNumberMapRootIndex,
instr->FalseLabel(chunk()),
DO_SMI_CHECK);
__ ldr(scratch, FieldMemOperand(value, HeapNumber::kExponentOffset));
__ ldr(ip, FieldMemOperand(value, HeapNumber::kMantissaOffset));
__ cmp(scratch, Operand(0x80000000));
__ cmp(ip, Operand(0x00000000), eq);
}
EmitBranch(instr, eq);
}
Condition LCodeGen::EmitIsObject(Register input,
Register temp1,
Label* is_not_object,
Label* is_object) {
Register temp2 = scratch0();
__ JumpIfSmi(input, is_not_object);
__ LoadRoot(temp2, Heap::kNullValueRootIndex);
__ cmp(input, temp2);
__ b(eq, is_object);
// Load map.
__ ldr(temp1, FieldMemOperand(input, HeapObject::kMapOffset));
// Undetectable objects behave like undefined.
__ ldrb(temp2, FieldMemOperand(temp1, Map::kBitFieldOffset));
__ tst(temp2, Operand(1 << Map::kIsUndetectable));
__ b(ne, is_not_object);
// Load instance type and check that it is in object type range.
__ ldrb(temp2, FieldMemOperand(temp1, Map::kInstanceTypeOffset));
__ cmp(temp2, Operand(FIRST_NONCALLABLE_SPEC_OBJECT_TYPE));
__ b(lt, is_not_object);
__ cmp(temp2, Operand(LAST_NONCALLABLE_SPEC_OBJECT_TYPE));
return le;
}
void LCodeGen::DoIsObjectAndBranch(LIsObjectAndBranch* instr) {
Register reg = ToRegister(instr->value());
Register temp1 = ToRegister(instr->temp());
Condition true_cond =
EmitIsObject(reg, temp1,
instr->FalseLabel(chunk_), instr->TrueLabel(chunk_));
EmitBranch(instr, true_cond);
}
Condition LCodeGen::EmitIsString(Register input,
Register temp1,
Label* is_not_string,
SmiCheck check_needed = INLINE_SMI_CHECK) {
if (check_needed == INLINE_SMI_CHECK) {
__ JumpIfSmi(input, is_not_string);
}
__ CompareObjectType(input, temp1, temp1, FIRST_NONSTRING_TYPE);
return lt;
}
void LCodeGen::DoIsStringAndBranch(LIsStringAndBranch* instr) {
Register reg = ToRegister(instr->value());
Register temp1 = ToRegister(instr->temp());
SmiCheck check_needed =
instr->hydrogen()->value()->type().IsHeapObject()
? OMIT_SMI_CHECK : INLINE_SMI_CHECK;
Condition true_cond =
EmitIsString(reg, temp1, instr->FalseLabel(chunk_), check_needed);
EmitBranch(instr, true_cond);
}
void LCodeGen::DoIsSmiAndBranch(LIsSmiAndBranch* instr) {
Register input_reg = EmitLoadRegister(instr->value(), ip);
__ SmiTst(input_reg);
EmitBranch(instr, eq);
}
void LCodeGen::DoIsUndetectableAndBranch(LIsUndetectableAndBranch* instr) {
Register input = ToRegister(instr->value());
Register temp = ToRegister(instr->temp());
if (!instr->hydrogen()->value()->type().IsHeapObject()) {
__ JumpIfSmi(input, instr->FalseLabel(chunk_));
}
__ ldr(temp, FieldMemOperand(input, HeapObject::kMapOffset));
__ ldrb(temp, FieldMemOperand(temp, Map::kBitFieldOffset));
__ tst(temp, Operand(1 << Map::kIsUndetectable));
EmitBranch(instr, ne);
}
static Condition ComputeCompareCondition(Token::Value op) {
switch (op) {
case Token::EQ_STRICT:
case Token::EQ:
return eq;
case Token::LT:
return lt;
case Token::GT:
return gt;
case Token::LTE:
return le;
case Token::GTE:
return ge;
default:
UNREACHABLE();
return kNoCondition;
}
}
void LCodeGen::DoStringCompareAndBranch(LStringCompareAndBranch* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
Token::Value op = instr->op();
Handle<Code> ic = CodeFactory::CompareIC(isolate(), op).code();
CallCode(ic, RelocInfo::CODE_TARGET, instr);
// This instruction also signals no smi code inlined.
__ cmp(r0, Operand::Zero());
Condition condition = ComputeCompareCondition(op);
EmitBranch(instr, condition);
}
static InstanceType TestType(HHasInstanceTypeAndBranch* instr) {
InstanceType from = instr->from();
InstanceType to = instr->to();
if (from == FIRST_TYPE) return to;
DCHECK(from == to || to == LAST_TYPE);
return from;
}
static Condition BranchCondition(HHasInstanceTypeAndBranch* instr) {
InstanceType from = instr->from();
InstanceType to = instr->to();
if (from == to) return eq;
if (to == LAST_TYPE) return hs;
if (from == FIRST_TYPE) return ls;
UNREACHABLE();
return eq;
}
void LCodeGen::DoHasInstanceTypeAndBranch(LHasInstanceTypeAndBranch* instr) {
Register scratch = scratch0();
Register input = ToRegister(instr->value());
if (!instr->hydrogen()->value()->type().IsHeapObject()) {
__ JumpIfSmi(input, instr->FalseLabel(chunk_));
}
__ CompareObjectType(input, scratch, scratch, TestType(instr->hydrogen()));
EmitBranch(instr, BranchCondition(instr->hydrogen()));
}
void LCodeGen::DoGetCachedArrayIndex(LGetCachedArrayIndex* instr) {
Register input = ToRegister(instr->value());
Register result = ToRegister(instr->result());
__ AssertString(input);
__ ldr(result, FieldMemOperand(input, String::kHashFieldOffset));
__ IndexFromHash(result, result);
}
void LCodeGen::DoHasCachedArrayIndexAndBranch(
LHasCachedArrayIndexAndBranch* instr) {
Register input = ToRegister(instr->value());
Register scratch = scratch0();
__ ldr(scratch,
FieldMemOperand(input, String::kHashFieldOffset));
__ tst(scratch, Operand(String::kContainsCachedArrayIndexMask));
EmitBranch(instr, eq);
}
// Branches to a label or falls through with the answer in flags. Trashes
// the temp registers, but not the input.
void LCodeGen::EmitClassOfTest(Label* is_true,
Label* is_false,
Handle<String>class_name,
Register input,
Register temp,
Register temp2) {
DCHECK(!input.is(temp));
DCHECK(!input.is(temp2));
DCHECK(!temp.is(temp2));
__ JumpIfSmi(input, is_false);
if (String::Equals(isolate()->factory()->Function_string(), class_name)) {
// Assuming the following assertions, we can use the same compares to test
// for both being a function type and being in the object type range.
STATIC_ASSERT(NUM_OF_CALLABLE_SPEC_OBJECT_TYPES == 2);
STATIC_ASSERT(FIRST_NONCALLABLE_SPEC_OBJECT_TYPE ==
FIRST_SPEC_OBJECT_TYPE + 1);
STATIC_ASSERT(LAST_NONCALLABLE_SPEC_OBJECT_TYPE ==
LAST_SPEC_OBJECT_TYPE - 1);
STATIC_ASSERT(LAST_SPEC_OBJECT_TYPE == LAST_TYPE);
__ CompareObjectType(input, temp, temp2, FIRST_SPEC_OBJECT_TYPE);
__ b(lt, is_false);
__ b(eq, is_true);
__ cmp(temp2, Operand(LAST_SPEC_OBJECT_TYPE));
__ b(eq, is_true);
} else {
// Faster code path to avoid two compares: subtract lower bound from the
// actual type and do a signed compare with the width of the type range.
__ ldr(temp, FieldMemOperand(input, HeapObject::kMapOffset));
__ ldrb(temp2, FieldMemOperand(temp, Map::kInstanceTypeOffset));
__ sub(temp2, temp2, Operand(FIRST_NONCALLABLE_SPEC_OBJECT_TYPE));
__ cmp(temp2, Operand(LAST_NONCALLABLE_SPEC_OBJECT_TYPE -
FIRST_NONCALLABLE_SPEC_OBJECT_TYPE));
__ b(gt, is_false);
}
// Now we are in the FIRST-LAST_NONCALLABLE_SPEC_OBJECT_TYPE range.
// Check if the constructor in the map is a function.
Register instance_type = ip;
__ GetMapConstructor(temp, temp, temp2, instance_type);
// Objects with a non-function constructor have class 'Object'.
__ cmp(instance_type, Operand(JS_FUNCTION_TYPE));
if (class_name->IsOneByteEqualTo(STATIC_CHAR_VECTOR("Object"))) {
__ b(ne, is_true);
} else {
__ b(ne, is_false);
}
// temp now contains the constructor function. Grab the
// instance class name from there.
__ ldr(temp, FieldMemOperand(temp, JSFunction::kSharedFunctionInfoOffset));
__ ldr(temp, FieldMemOperand(temp,
SharedFunctionInfo::kInstanceClassNameOffset));
// The class name we are testing against is internalized since it's a literal.
// The name in the constructor is internalized because of the way the context
// is booted. This routine isn't expected to work for random API-created
// classes and it doesn't have to because you can't access it with natives
// syntax. Since both sides are internalized it is sufficient to use an
// identity comparison.
__ cmp(temp, Operand(class_name));
// End with the answer in flags.
}
void LCodeGen::DoClassOfTestAndBranch(LClassOfTestAndBranch* instr) {
Register input = ToRegister(instr->value());
Register temp = scratch0();
Register temp2 = ToRegister(instr->temp());
Handle<String> class_name = instr->hydrogen()->class_name();
EmitClassOfTest(instr->TrueLabel(chunk_), instr->FalseLabel(chunk_),
class_name, input, temp, temp2);
EmitBranch(instr, eq);
}
void LCodeGen::DoCmpMapAndBranch(LCmpMapAndBranch* instr) {
Register reg = ToRegister(instr->value());
Register temp = ToRegister(instr->temp());
__ ldr(temp, FieldMemOperand(reg, HeapObject::kMapOffset));
__ cmp(temp, Operand(instr->map()));
EmitBranch(instr, eq);
}
void LCodeGen::DoInstanceOf(LInstanceOf* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->left()).is(r0)); // Object is in r0.
DCHECK(ToRegister(instr->right()).is(r1)); // Function is in r1.
InstanceofStub stub(isolate(), InstanceofStub::kArgsInRegisters);
CallCode(stub.GetCode(), RelocInfo::CODE_TARGET, instr);
__ cmp(r0, Operand::Zero());
__ mov(r0, Operand(factory()->false_value()), LeaveCC, ne);
__ mov(r0, Operand(factory()->true_value()), LeaveCC, eq);
}
void LCodeGen::DoInstanceOfKnownGlobal(LInstanceOfKnownGlobal* instr) {
class DeferredInstanceOfKnownGlobal FINAL : public LDeferredCode {
public:
DeferredInstanceOfKnownGlobal(LCodeGen* codegen,
LInstanceOfKnownGlobal* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE {
[Arm]: Enable use of extended out-of-line constant pool for Arm. - Adds support to the Arm assembler to use extended constant pools. - Update (set_)target_address_at to support extended constant pool load updates. - Replace Operand::is_single_instruction with Operand::instructions_required Due to the fact that different constant pool load types require different numbers of instructions. - Various cleanups of ConstantPoolBuilder to cleaner integration of the extended constant pool building. - Update GetRelocatedValue such that offsets to both map_check and bool_load are explicitly provided, rather than location of bool_load being inferred based on map_check, since the code inbetween is no longer of a predictable size. - Update MacroAssembler::GetRelocatedValueLocation() to add support for getting a value from an extended constant pool entry. - Update Debug::SetAfterBreakTarget() to use target_address_from_return_address when checking for debug breaks at constant pool load points. - Change ConstantPoolIterateBody to iterate over both heap and code pointer in the small section before moving onto the extended section, to work around the requirement of the serializer that pointers are iterated in-order. - Increase old_pointer_space SizeOfFirstPage() to offset the fact that constant pools are now in the old pointer space (rather than code). R=ulan@chromium.org Review URL: https://codereview.chromium.org/356393003 git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22209 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-07-03 17:01:14 +00:00
codegen()->DoDeferredInstanceOfKnownGlobal(instr_, &map_check_,
&load_bool_);
}
LInstruction* instr() OVERRIDE { return instr_; }
Label* map_check() { return &map_check_; }
[Arm]: Enable use of extended out-of-line constant pool for Arm. - Adds support to the Arm assembler to use extended constant pools. - Update (set_)target_address_at to support extended constant pool load updates. - Replace Operand::is_single_instruction with Operand::instructions_required Due to the fact that different constant pool load types require different numbers of instructions. - Various cleanups of ConstantPoolBuilder to cleaner integration of the extended constant pool building. - Update GetRelocatedValue such that offsets to both map_check and bool_load are explicitly provided, rather than location of bool_load being inferred based on map_check, since the code inbetween is no longer of a predictable size. - Update MacroAssembler::GetRelocatedValueLocation() to add support for getting a value from an extended constant pool entry. - Update Debug::SetAfterBreakTarget() to use target_address_from_return_address when checking for debug breaks at constant pool load points. - Change ConstantPoolIterateBody to iterate over both heap and code pointer in the small section before moving onto the extended section, to work around the requirement of the serializer that pointers are iterated in-order. - Increase old_pointer_space SizeOfFirstPage() to offset the fact that constant pools are now in the old pointer space (rather than code). R=ulan@chromium.org Review URL: https://codereview.chromium.org/356393003 git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22209 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-07-03 17:01:14 +00:00
Label* load_bool() { return &load_bool_; }
private:
LInstanceOfKnownGlobal* instr_;
Label map_check_;
[Arm]: Enable use of extended out-of-line constant pool for Arm. - Adds support to the Arm assembler to use extended constant pools. - Update (set_)target_address_at to support extended constant pool load updates. - Replace Operand::is_single_instruction with Operand::instructions_required Due to the fact that different constant pool load types require different numbers of instructions. - Various cleanups of ConstantPoolBuilder to cleaner integration of the extended constant pool building. - Update GetRelocatedValue such that offsets to both map_check and bool_load are explicitly provided, rather than location of bool_load being inferred based on map_check, since the code inbetween is no longer of a predictable size. - Update MacroAssembler::GetRelocatedValueLocation() to add support for getting a value from an extended constant pool entry. - Update Debug::SetAfterBreakTarget() to use target_address_from_return_address when checking for debug breaks at constant pool load points. - Change ConstantPoolIterateBody to iterate over both heap and code pointer in the small section before moving onto the extended section, to work around the requirement of the serializer that pointers are iterated in-order. - Increase old_pointer_space SizeOfFirstPage() to offset the fact that constant pools are now in the old pointer space (rather than code). R=ulan@chromium.org Review URL: https://codereview.chromium.org/356393003 git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22209 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-07-03 17:01:14 +00:00
Label load_bool_;
};
DeferredInstanceOfKnownGlobal* deferred;
deferred = new(zone()) DeferredInstanceOfKnownGlobal(this, instr);
Label done, false_result;
Register object = ToRegister(instr->value());
Register temp = ToRegister(instr->temp());
Register result = ToRegister(instr->result());
// A Smi is not instance of anything.
__ JumpIfSmi(object, &false_result);
// This is the inlined call site instanceof cache. The two occurences of the
// hole value will be patched to the last map/result pair generated by the
// instanceof stub.
Label cache_miss;
Register map = temp;
__ ldr(map, FieldMemOperand(object, HeapObject::kMapOffset));
{
// Block constant pool emission to ensure the positions of instructions are
// as expected by the patcher. See InstanceofStub::Generate().
Assembler::BlockConstPoolScope block_const_pool(masm());
__ bind(deferred->map_check()); // Label for calculating code patching.
// We use Factory::the_hole_value() on purpose instead of loading from the
// root array to force relocation to be able to later patch with
// the cached map.
Handle<Cell> cell = factory()->NewCell(factory()->the_hole_value());
__ mov(ip, Operand(cell));
__ ldr(ip, FieldMemOperand(ip, Cell::kValueOffset));
__ cmp(map, Operand(ip));
__ b(ne, &cache_miss);
[Arm]: Enable use of extended out-of-line constant pool for Arm. - Adds support to the Arm assembler to use extended constant pools. - Update (set_)target_address_at to support extended constant pool load updates. - Replace Operand::is_single_instruction with Operand::instructions_required Due to the fact that different constant pool load types require different numbers of instructions. - Various cleanups of ConstantPoolBuilder to cleaner integration of the extended constant pool building. - Update GetRelocatedValue such that offsets to both map_check and bool_load are explicitly provided, rather than location of bool_load being inferred based on map_check, since the code inbetween is no longer of a predictable size. - Update MacroAssembler::GetRelocatedValueLocation() to add support for getting a value from an extended constant pool entry. - Update Debug::SetAfterBreakTarget() to use target_address_from_return_address when checking for debug breaks at constant pool load points. - Change ConstantPoolIterateBody to iterate over both heap and code pointer in the small section before moving onto the extended section, to work around the requirement of the serializer that pointers are iterated in-order. - Increase old_pointer_space SizeOfFirstPage() to offset the fact that constant pools are now in the old pointer space (rather than code). R=ulan@chromium.org Review URL: https://codereview.chromium.org/356393003 git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22209 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-07-03 17:01:14 +00:00
__ bind(deferred->load_bool()); // Label for calculating code patching.
// We use Factory::the_hole_value() on purpose instead of loading from the
// root array to force relocation to be able to later patch
// with true or false.
__ mov(result, Operand(factory()->the_hole_value()));
}
__ b(&done);
// The inlined call site cache did not match. Check null and string before
// calling the deferred code.
__ bind(&cache_miss);
// Null is not instance of anything.
__ LoadRoot(ip, Heap::kNullValueRootIndex);
__ cmp(object, Operand(ip));
__ b(eq, &false_result);
// String values is not instance of anything.
Condition is_string = masm_->IsObjectStringType(object, temp);
__ b(is_string, &false_result);
// Go to the deferred code.
__ b(deferred->entry());
__ bind(&false_result);
__ LoadRoot(result, Heap::kFalseValueRootIndex);
// Here result has either true or false. Deferred code also produces true or
// false object.
__ bind(deferred->exit());
__ bind(&done);
}
void LCodeGen::DoDeferredInstanceOfKnownGlobal(LInstanceOfKnownGlobal* instr,
[Arm]: Enable use of extended out-of-line constant pool for Arm. - Adds support to the Arm assembler to use extended constant pools. - Update (set_)target_address_at to support extended constant pool load updates. - Replace Operand::is_single_instruction with Operand::instructions_required Due to the fact that different constant pool load types require different numbers of instructions. - Various cleanups of ConstantPoolBuilder to cleaner integration of the extended constant pool building. - Update GetRelocatedValue such that offsets to both map_check and bool_load are explicitly provided, rather than location of bool_load being inferred based on map_check, since the code inbetween is no longer of a predictable size. - Update MacroAssembler::GetRelocatedValueLocation() to add support for getting a value from an extended constant pool entry. - Update Debug::SetAfterBreakTarget() to use target_address_from_return_address when checking for debug breaks at constant pool load points. - Change ConstantPoolIterateBody to iterate over both heap and code pointer in the small section before moving onto the extended section, to work around the requirement of the serializer that pointers are iterated in-order. - Increase old_pointer_space SizeOfFirstPage() to offset the fact that constant pools are now in the old pointer space (rather than code). R=ulan@chromium.org Review URL: https://codereview.chromium.org/356393003 git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22209 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-07-03 17:01:14 +00:00
Label* map_check,
Label* bool_load) {
InstanceofStub::Flags flags = InstanceofStub::kNoFlags;
flags = static_cast<InstanceofStub::Flags>(
flags | InstanceofStub::kArgsInRegisters);
flags = static_cast<InstanceofStub::Flags>(
flags | InstanceofStub::kCallSiteInlineCheck);
flags = static_cast<InstanceofStub::Flags>(
flags | InstanceofStub::kReturnTrueFalseObject);
InstanceofStub stub(isolate(), flags);
PushSafepointRegistersScope scope(this);
LoadContextFromDeferred(instr->context());
__ Move(InstanceofStub::right(), instr->function());
[Arm]: Enable use of extended out-of-line constant pool for Arm. - Adds support to the Arm assembler to use extended constant pools. - Update (set_)target_address_at to support extended constant pool load updates. - Replace Operand::is_single_instruction with Operand::instructions_required Due to the fact that different constant pool load types require different numbers of instructions. - Various cleanups of ConstantPoolBuilder to cleaner integration of the extended constant pool building. - Update GetRelocatedValue such that offsets to both map_check and bool_load are explicitly provided, rather than location of bool_load being inferred based on map_check, since the code inbetween is no longer of a predictable size. - Update MacroAssembler::GetRelocatedValueLocation() to add support for getting a value from an extended constant pool entry. - Update Debug::SetAfterBreakTarget() to use target_address_from_return_address when checking for debug breaks at constant pool load points. - Change ConstantPoolIterateBody to iterate over both heap and code pointer in the small section before moving onto the extended section, to work around the requirement of the serializer that pointers are iterated in-order. - Increase old_pointer_space SizeOfFirstPage() to offset the fact that constant pools are now in the old pointer space (rather than code). R=ulan@chromium.org Review URL: https://codereview.chromium.org/356393003 git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22209 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-07-03 17:01:14 +00:00
int call_size = CallCodeSize(stub.GetCode(), RelocInfo::CODE_TARGET);
int additional_delta = (call_size / Assembler::kInstrSize) + 4;
// Make sure that code size is predicable, since we use specific constants
// offsets in the code to find embedded values..
[Arm]: Enable use of extended out-of-line constant pool for Arm. - Adds support to the Arm assembler to use extended constant pools. - Update (set_)target_address_at to support extended constant pool load updates. - Replace Operand::is_single_instruction with Operand::instructions_required Due to the fact that different constant pool load types require different numbers of instructions. - Various cleanups of ConstantPoolBuilder to cleaner integration of the extended constant pool building. - Update GetRelocatedValue such that offsets to both map_check and bool_load are explicitly provided, rather than location of bool_load being inferred based on map_check, since the code inbetween is no longer of a predictable size. - Update MacroAssembler::GetRelocatedValueLocation() to add support for getting a value from an extended constant pool entry. - Update Debug::SetAfterBreakTarget() to use target_address_from_return_address when checking for debug breaks at constant pool load points. - Change ConstantPoolIterateBody to iterate over both heap and code pointer in the small section before moving onto the extended section, to work around the requirement of the serializer that pointers are iterated in-order. - Increase old_pointer_space SizeOfFirstPage() to offset the fact that constant pools are now in the old pointer space (rather than code). R=ulan@chromium.org Review URL: https://codereview.chromium.org/356393003 git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22209 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-07-03 17:01:14 +00:00
PredictableCodeSizeScope predictable(
masm_, (additional_delta + 1) * Assembler::kInstrSize);
// Make sure we don't emit any additional entries in the constant pool before
// the call to ensure that the CallCodeSize() calculated the correct number of
// instructions for the constant pool load.
{
ConstantPoolUnavailableScope constant_pool_unavailable(masm_);
int map_check_delta =
masm_->InstructionsGeneratedSince(map_check) + additional_delta;
int bool_load_delta =
masm_->InstructionsGeneratedSince(bool_load) + additional_delta;
Label before_push_delta;
__ bind(&before_push_delta);
__ BlockConstPoolFor(additional_delta);
// r5 is used to communicate the offset to the location of the map check.
__ mov(r5, Operand(map_check_delta * kPointerSize));
// r6 is used to communicate the offset to the location of the bool load.
__ mov(r6, Operand(bool_load_delta * kPointerSize));
// The mov above can generate one or two instructions. The delta was
// computed for two instructions, so we need to pad here in case of one
// instruction.
while (masm_->InstructionsGeneratedSince(&before_push_delta) != 4) {
__ nop();
}
}
CallCodeGeneric(stub.GetCode(),
RelocInfo::CODE_TARGET,
instr,
RECORD_SAFEPOINT_WITH_REGISTERS_AND_NO_ARGUMENTS);
LEnvironment* env = instr->GetDeferredLazyDeoptimizationEnvironment();
safepoints_.RecordLazyDeoptimizationIndex(env->deoptimization_index());
// Put the result value (r0) into the result register slot and
// restore all registers.
__ StoreToSafepointRegisterSlot(r0, ToRegister(instr->result()));
}
void LCodeGen::DoCmpT(LCmpT* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
Token::Value op = instr->op();
Handle<Code> ic = CodeFactory::CompareIC(isolate(), op).code();
CallCode(ic, RelocInfo::CODE_TARGET, instr);
// This instruction also signals no smi code inlined.
__ cmp(r0, Operand::Zero());
Condition condition = ComputeCompareCondition(op);
__ LoadRoot(ToRegister(instr->result()),
Heap::kTrueValueRootIndex,
condition);
__ LoadRoot(ToRegister(instr->result()),
Heap::kFalseValueRootIndex,
NegateCondition(condition));
}
void LCodeGen::DoReturn(LReturn* instr) {
if (FLAG_trace && info()->IsOptimizing()) {
// Push the return value on the stack as the parameter.
// Runtime::TraceExit returns its parameter in r0. We're leaving the code
// managed by the register allocator and tearing down the frame, it's
// safe to write to the context register.
__ push(r0);
__ ldr(cp, MemOperand(fp, StandardFrameConstants::kContextOffset));
__ CallRuntime(Runtime::kTraceExit, 1);
}
if (info()->saves_caller_doubles()) {
RestoreCallerDoubles();
}
int no_frame_start = -1;
if (NeedsEagerFrame()) {
no_frame_start = masm_->LeaveFrame(StackFrame::JAVA_SCRIPT);
}
{ ConstantPoolUnavailableScope constant_pool_unavailable(masm());
if (instr->has_constant_parameter_count()) {
int parameter_count = ToInteger32(instr->constant_parameter_count());
int32_t sp_delta = (parameter_count + 1) * kPointerSize;
if (sp_delta != 0) {
__ add(sp, sp, Operand(sp_delta));
}
} else {
DCHECK(info()->IsStub()); // Functions would need to drop one more value.
Register reg = ToRegister(instr->parameter_count());
// The argument count parameter is a smi
__ SmiUntag(reg);
__ add(sp, sp, Operand(reg, LSL, kPointerSizeLog2));
}
__ Jump(lr);
if (no_frame_start != -1) {
info_->AddNoFrameRange(no_frame_start, masm_->pc_offset());
}
}
}
template <class T>
void LCodeGen::EmitVectorLoadICRegisters(T* instr) {
DCHECK(FLAG_vector_ics);
Register vector_register = ToRegister(instr->temp_vector());
Register slot_register = VectorLoadICDescriptor::SlotRegister();
DCHECK(vector_register.is(VectorLoadICDescriptor::VectorRegister()));
DCHECK(slot_register.is(r0));
AllowDeferredHandleDereference vector_structure_check;
Handle<TypeFeedbackVector> vector = instr->hydrogen()->feedback_vector();
__ Move(vector_register, vector);
// No need to allocate this register.
FeedbackVectorICSlot slot = instr->hydrogen()->slot();
int index = vector->GetIndex(slot);
__ mov(slot_register, Operand(Smi::FromInt(index)));
}
void LCodeGen::DoLoadGlobalGeneric(LLoadGlobalGeneric* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->global_object())
.is(LoadDescriptor::ReceiverRegister()));
DCHECK(ToRegister(instr->result()).is(r0));
__ mov(LoadDescriptor::NameRegister(), Operand(instr->name()));
if (FLAG_vector_ics) {
EmitVectorLoadICRegisters<LLoadGlobalGeneric>(instr);
}
ContextualMode mode = instr->for_typeof() ? NOT_CONTEXTUAL : CONTEXTUAL;
Handle<Code> ic = CodeFactory::LoadICInOptimizedCode(isolate(), mode,
PREMONOMORPHIC).code();
CallCode(ic, RelocInfo::CODE_TARGET, instr);
}
void LCodeGen::DoLoadContextSlot(LLoadContextSlot* instr) {
Register context = ToRegister(instr->context());
Register result = ToRegister(instr->result());
__ ldr(result, ContextOperand(context, instr->slot_index()));
if (instr->hydrogen()->RequiresHoleCheck()) {
__ LoadRoot(ip, Heap::kTheHoleValueRootIndex);
__ cmp(result, ip);
if (instr->hydrogen()->DeoptimizesOnHole()) {
DeoptimizeIf(eq, instr, Deoptimizer::kHole);
} else {
__ mov(result, Operand(factory()->undefined_value()), LeaveCC, eq);
}
}
}
void LCodeGen::DoStoreContextSlot(LStoreContextSlot* instr) {
Register context = ToRegister(instr->context());
Register value = ToRegister(instr->value());
Register scratch = scratch0();
MemOperand target = ContextOperand(context, instr->slot_index());
Label skip_assignment;
if (instr->hydrogen()->RequiresHoleCheck()) {
__ ldr(scratch, target);
__ LoadRoot(ip, Heap::kTheHoleValueRootIndex);
__ cmp(scratch, ip);
if (instr->hydrogen()->DeoptimizesOnHole()) {
DeoptimizeIf(eq, instr, Deoptimizer::kHole);
} else {
__ b(ne, &skip_assignment);
}
}
__ str(value, target);
if (instr->hydrogen()->NeedsWriteBarrier()) {
SmiCheck check_needed =
instr->hydrogen()->value()->type().IsHeapObject()
? OMIT_SMI_CHECK : INLINE_SMI_CHECK;
__ RecordWriteContextSlot(context,
target.offset(),
value,
scratch,
GetLinkRegisterState(),
kSaveFPRegs,
EMIT_REMEMBERED_SET,
check_needed);
}
__ bind(&skip_assignment);
}
void LCodeGen::DoLoadNamedField(LLoadNamedField* instr) {
HObjectAccess access = instr->hydrogen()->access();
int offset = access.offset();
Register object = ToRegister(instr->object());
if (access.IsExternalMemory()) {
Register result = ToRegister(instr->result());
MemOperand operand = MemOperand(object, offset);
__ Load(result, operand, access.representation());
return;
}
if (instr->hydrogen()->representation().IsDouble()) {
DwVfpRegister result = ToDoubleRegister(instr->result());
__ vldr(result, FieldMemOperand(object, offset));
return;
}
Register result = ToRegister(instr->result());
if (!access.IsInobject()) {
__ ldr(result, FieldMemOperand(object, JSObject::kPropertiesOffset));
object = result;
}
MemOperand operand = FieldMemOperand(object, offset);
__ Load(result, operand, access.representation());
}
void LCodeGen::DoLoadNamedGeneric(LLoadNamedGeneric* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->object()).is(LoadDescriptor::ReceiverRegister()));
DCHECK(ToRegister(instr->result()).is(r0));
// Name is always in r2.
__ mov(LoadDescriptor::NameRegister(), Operand(instr->name()));
if (FLAG_vector_ics) {
EmitVectorLoadICRegisters<LLoadNamedGeneric>(instr);
}
Handle<Code> ic = CodeFactory::LoadICInOptimizedCode(
isolate(), NOT_CONTEXTUAL,
instr->hydrogen()->initialization_state()).code();
CallCode(ic, RelocInfo::CODE_TARGET, instr, NEVER_INLINE_TARGET_ADDRESS);
}
void LCodeGen::DoLoadFunctionPrototype(LLoadFunctionPrototype* instr) {
Register scratch = scratch0();
Register function = ToRegister(instr->function());
Register result = ToRegister(instr->result());
// Get the prototype or initial map from the function.
__ ldr(result,
FieldMemOperand(function, JSFunction::kPrototypeOrInitialMapOffset));
// Check that the function has a prototype or an initial map.
__ LoadRoot(ip, Heap::kTheHoleValueRootIndex);
__ cmp(result, ip);
DeoptimizeIf(eq, instr, Deoptimizer::kHole);
// If the function does not have an initial map, we're done.
Label done;
__ CompareObjectType(result, scratch, scratch, MAP_TYPE);
__ b(ne, &done);
// Get the prototype from the initial map.
__ ldr(result, FieldMemOperand(result, Map::kPrototypeOffset));
// All done.
__ bind(&done);
}
void LCodeGen::DoLoadRoot(LLoadRoot* instr) {
Register result = ToRegister(instr->result());
__ LoadRoot(result, instr->index());
}
void LCodeGen::DoAccessArgumentsAt(LAccessArgumentsAt* instr) {
Register arguments = ToRegister(instr->arguments());
Register result = ToRegister(instr->result());
// There are two words between the frame pointer and the last argument.
// Subtracting from length accounts for one of them add one more.
if (instr->length()->IsConstantOperand()) {
int const_length = ToInteger32(LConstantOperand::cast(instr->length()));
if (instr->index()->IsConstantOperand()) {
int const_index = ToInteger32(LConstantOperand::cast(instr->index()));
int index = (const_length - const_index) + 1;
__ ldr(result, MemOperand(arguments, index * kPointerSize));
} else {
Register index = ToRegister(instr->index());
__ rsb(result, index, Operand(const_length + 1));
__ ldr(result, MemOperand(arguments, result, LSL, kPointerSizeLog2));
}
} else if (instr->index()->IsConstantOperand()) {
Register length = ToRegister(instr->length());
int const_index = ToInteger32(LConstantOperand::cast(instr->index()));
int loc = const_index - 1;
if (loc != 0) {
__ sub(result, length, Operand(loc));
__ ldr(result, MemOperand(arguments, result, LSL, kPointerSizeLog2));
} else {
__ ldr(result, MemOperand(arguments, length, LSL, kPointerSizeLog2));
}
} else {
Register length = ToRegister(instr->length());
Register index = ToRegister(instr->index());
__ sub(result, length, index);
__ add(result, result, Operand(1));
__ ldr(result, MemOperand(arguments, result, LSL, kPointerSizeLog2));
}
}
void LCodeGen::DoLoadKeyedExternalArray(LLoadKeyed* instr) {
Register external_pointer = ToRegister(instr->elements());
Register key = no_reg;
ElementsKind elements_kind = instr->elements_kind();
bool key_is_constant = instr->key()->IsConstantOperand();
int constant_key = 0;
if (key_is_constant) {
constant_key = ToInteger32(LConstantOperand::cast(instr->key()));
if (constant_key & 0xF0000000) {
Abort(kArrayIndexConstantValueTooBig);
}
} else {
key = ToRegister(instr->key());
}
int element_size_shift = ElementsKindToShiftSize(elements_kind);
int shift_size = (instr->hydrogen()->key()->representation().IsSmi())
? (element_size_shift - kSmiTagSize) : element_size_shift;
int base_offset = instr->base_offset();
if (elements_kind == EXTERNAL_FLOAT32_ELEMENTS ||
elements_kind == FLOAT32_ELEMENTS ||
elements_kind == EXTERNAL_FLOAT64_ELEMENTS ||
elements_kind == FLOAT64_ELEMENTS) {
int base_offset = instr->base_offset();
DwVfpRegister result = ToDoubleRegister(instr->result());
Operand operand = key_is_constant
? Operand(constant_key << element_size_shift)
: Operand(key, LSL, shift_size);
__ add(scratch0(), external_pointer, operand);
if (elements_kind == EXTERNAL_FLOAT32_ELEMENTS ||
elements_kind == FLOAT32_ELEMENTS) {
__ vldr(double_scratch0().low(), scratch0(), base_offset);
__ vcvt_f64_f32(result, double_scratch0().low());
} else { // i.e. elements_kind == EXTERNAL_DOUBLE_ELEMENTS
__ vldr(result, scratch0(), base_offset);
}
} else {
Register result = ToRegister(instr->result());
MemOperand mem_operand = PrepareKeyedOperand(
key, external_pointer, key_is_constant, constant_key,
element_size_shift, shift_size, base_offset);
switch (elements_kind) {
case EXTERNAL_INT8_ELEMENTS:
case INT8_ELEMENTS:
__ ldrsb(result, mem_operand);
break;
case EXTERNAL_UINT8_CLAMPED_ELEMENTS:
case EXTERNAL_UINT8_ELEMENTS:
case UINT8_ELEMENTS:
case UINT8_CLAMPED_ELEMENTS:
__ ldrb(result, mem_operand);
break;
case EXTERNAL_INT16_ELEMENTS:
case INT16_ELEMENTS:
__ ldrsh(result, mem_operand);
break;
case EXTERNAL_UINT16_ELEMENTS:
case UINT16_ELEMENTS:
__ ldrh(result, mem_operand);
break;
case EXTERNAL_INT32_ELEMENTS:
case INT32_ELEMENTS:
__ ldr(result, mem_operand);
break;
case EXTERNAL_UINT32_ELEMENTS:
case UINT32_ELEMENTS:
__ ldr(result, mem_operand);
if (!instr->hydrogen()->CheckFlag(HInstruction::kUint32)) {
__ cmp(result, Operand(0x80000000));
DeoptimizeIf(cs, instr, Deoptimizer::kNegativeValue);
}
break;
case FLOAT32_ELEMENTS:
case FLOAT64_ELEMENTS:
case EXTERNAL_FLOAT32_ELEMENTS:
case EXTERNAL_FLOAT64_ELEMENTS:
case FAST_HOLEY_DOUBLE_ELEMENTS:
case FAST_HOLEY_ELEMENTS:
case FAST_HOLEY_SMI_ELEMENTS:
case FAST_DOUBLE_ELEMENTS:
case FAST_ELEMENTS:
case FAST_SMI_ELEMENTS:
case DICTIONARY_ELEMENTS:
case SLOPPY_ARGUMENTS_ELEMENTS:
UNREACHABLE();
break;
}
}
}
void LCodeGen::DoLoadKeyedFixedDoubleArray(LLoadKeyed* instr) {
Register elements = ToRegister(instr->elements());
bool key_is_constant = instr->key()->IsConstantOperand();
Register key = no_reg;
DwVfpRegister result = ToDoubleRegister(instr->result());
Register scratch = scratch0();
int element_size_shift = ElementsKindToShiftSize(FAST_DOUBLE_ELEMENTS);
int base_offset = instr->base_offset();
if (key_is_constant) {
int constant_key = ToInteger32(LConstantOperand::cast(instr->key()));
if (constant_key & 0xF0000000) {
Abort(kArrayIndexConstantValueTooBig);
}
base_offset += constant_key * kDoubleSize;
}
__ add(scratch, elements, Operand(base_offset));
if (!key_is_constant) {
key = ToRegister(instr->key());
int shift_size = (instr->hydrogen()->key()->representation().IsSmi())
? (element_size_shift - kSmiTagSize) : element_size_shift;
__ add(scratch, scratch, Operand(key, LSL, shift_size));
}
__ vldr(result, scratch, 0);
if (instr->hydrogen()->RequiresHoleCheck()) {
__ ldr(scratch, MemOperand(scratch, sizeof(kHoleNanLower32)));
__ cmp(scratch, Operand(kHoleNanUpper32));
DeoptimizeIf(eq, instr, Deoptimizer::kHole);
}
}
void LCodeGen::DoLoadKeyedFixedArray(LLoadKeyed* instr) {
Register elements = ToRegister(instr->elements());
Register result = ToRegister(instr->result());
Register scratch = scratch0();
Register store_base = scratch;
int offset = instr->base_offset();
if (instr->key()->IsConstantOperand()) {
LConstantOperand* const_operand = LConstantOperand::cast(instr->key());
offset += ToInteger32(const_operand) * kPointerSize;
store_base = elements;
} else {
Register key = ToRegister(instr->key());
// Even though the HLoadKeyed instruction forces the input
// representation for the key to be an integer, the input gets replaced
// during bound check elimination with the index argument to the bounds
// check, which can be tagged, so that case must be handled here, too.
if (instr->hydrogen()->key()->representation().IsSmi()) {
__ add(scratch, elements, Operand::PointerOffsetFromSmiKey(key));
} else {
__ add(scratch, elements, Operand(key, LSL, kPointerSizeLog2));
}
}
__ ldr(result, MemOperand(store_base, offset));
// Check for the hole value.
if (instr->hydrogen()->RequiresHoleCheck()) {
if (IsFastSmiElementsKind(instr->hydrogen()->elements_kind())) {
__ SmiTst(result);
DeoptimizeIf(ne, instr, Deoptimizer::kNotASmi);
} else {
__ LoadRoot(scratch, Heap::kTheHoleValueRootIndex);
__ cmp(result, scratch);
DeoptimizeIf(eq, instr, Deoptimizer::kHole);
}
}
}
void LCodeGen::DoLoadKeyed(LLoadKeyed* instr) {
if (instr->is_typed_elements()) {
DoLoadKeyedExternalArray(instr);
} else if (instr->hydrogen()->representation().IsDouble()) {
DoLoadKeyedFixedDoubleArray(instr);
} else {
DoLoadKeyedFixedArray(instr);
}
}
MemOperand LCodeGen::PrepareKeyedOperand(Register key,
Register base,
bool key_is_constant,
int constant_key,
int element_size,
int shift_size,
int base_offset) {
if (key_is_constant) {
return MemOperand(base, (constant_key << element_size) + base_offset);
}
if (base_offset == 0) {
if (shift_size >= 0) {
return MemOperand(base, key, LSL, shift_size);
} else {
DCHECK_EQ(-1, shift_size);
return MemOperand(base, key, LSR, 1);
}
}
if (shift_size >= 0) {
__ add(scratch0(), base, Operand(key, LSL, shift_size));
return MemOperand(scratch0(), base_offset);
} else {
DCHECK_EQ(-1, shift_size);
__ add(scratch0(), base, Operand(key, ASR, 1));
return MemOperand(scratch0(), base_offset);
}
}
void LCodeGen::DoLoadKeyedGeneric(LLoadKeyedGeneric* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->object()).is(LoadDescriptor::ReceiverRegister()));
DCHECK(ToRegister(instr->key()).is(LoadDescriptor::NameRegister()));
if (FLAG_vector_ics) {
EmitVectorLoadICRegisters<LLoadKeyedGeneric>(instr);
}
Handle<Code> ic =
CodeFactory::KeyedLoadICInOptimizedCode(
isolate(), instr->hydrogen()->initialization_state()).code();
CallCode(ic, RelocInfo::CODE_TARGET, instr, NEVER_INLINE_TARGET_ADDRESS);
}
void LCodeGen::DoArgumentsElements(LArgumentsElements* instr) {
Register scratch = scratch0();
Register result = ToRegister(instr->result());
if (instr->hydrogen()->from_inlined()) {
__ sub(result, sp, Operand(2 * kPointerSize));
} else {
// Check if the calling frame is an arguments adaptor frame.
Label done, adapted;
__ ldr(scratch, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
__ ldr(result, MemOperand(scratch, StandardFrameConstants::kContextOffset));
__ cmp(result, Operand(Smi::FromInt(StackFrame::ARGUMENTS_ADAPTOR)));
// Result is the frame pointer for the frame if not adapted and for the real
// frame below the adaptor frame if adapted.
__ mov(result, fp, LeaveCC, ne);
__ mov(result, scratch, LeaveCC, eq);
}
}
void LCodeGen::DoArgumentsLength(LArgumentsLength* instr) {
Register elem = ToRegister(instr->elements());
Register result = ToRegister(instr->result());
Label done;
// If no arguments adaptor frame the number of arguments is fixed.
__ cmp(fp, elem);
__ mov(result, Operand(scope()->num_parameters()));
__ b(eq, &done);
// Arguments adaptor frame present. Get argument length from there.
__ ldr(result, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
__ ldr(result,
MemOperand(result, ArgumentsAdaptorFrameConstants::kLengthOffset));
__ SmiUntag(result);
// Argument length is in result register.
__ bind(&done);
}
void LCodeGen::DoWrapReceiver(LWrapReceiver* instr) {
Register receiver = ToRegister(instr->receiver());
Register function = ToRegister(instr->function());
Register result = ToRegister(instr->result());
Register scratch = scratch0();
// If the receiver is null or undefined, we have to pass the global
// object as a receiver to normal functions. Values have to be
// passed unchanged to builtins and strict-mode functions.
Label global_object, result_in_receiver;
if (!instr->hydrogen()->known_function()) {
// Do not transform the receiver to object for strict mode
// functions.
__ ldr(scratch,
FieldMemOperand(function, JSFunction::kSharedFunctionInfoOffset));
__ ldr(scratch,
FieldMemOperand(scratch, SharedFunctionInfo::kCompilerHintsOffset));
int mask = 1 << (SharedFunctionInfo::kStrictModeFunction + kSmiTagSize);
__ tst(scratch, Operand(mask));
__ b(ne, &result_in_receiver);
// Do not transform the receiver to object for builtins.
__ tst(scratch, Operand(1 << (SharedFunctionInfo::kNative + kSmiTagSize)));
__ b(ne, &result_in_receiver);
}
// Normal function. Replace undefined or null with global receiver.
__ LoadRoot(scratch, Heap::kNullValueRootIndex);
__ cmp(receiver, scratch);
__ b(eq, &global_object);
__ LoadRoot(scratch, Heap::kUndefinedValueRootIndex);
__ cmp(receiver, scratch);
__ b(eq, &global_object);
// Deoptimize if the receiver is not a JS object.
__ SmiTst(receiver);
DeoptimizeIf(eq, instr, Deoptimizer::kSmi);
__ CompareObjectType(receiver, scratch, scratch, FIRST_SPEC_OBJECT_TYPE);
DeoptimizeIf(lt, instr, Deoptimizer::kNotAJavaScriptObject);
__ b(&result_in_receiver);
__ bind(&global_object);
__ ldr(result, FieldMemOperand(function, JSFunction::kContextOffset));
__ ldr(result,
ContextOperand(result, Context::GLOBAL_OBJECT_INDEX));
__ ldr(result, FieldMemOperand(result, GlobalObject::kGlobalProxyOffset));
if (result.is(receiver)) {
__ bind(&result_in_receiver);
} else {
Label result_ok;
__ b(&result_ok);
__ bind(&result_in_receiver);
__ mov(result, receiver);
__ bind(&result_ok);
}
}
void LCodeGen::DoApplyArguments(LApplyArguments* instr) {
Register receiver = ToRegister(instr->receiver());
Register function = ToRegister(instr->function());
Register length = ToRegister(instr->length());
Register elements = ToRegister(instr->elements());
Register scratch = scratch0();
DCHECK(receiver.is(r0)); // Used for parameter count.
DCHECK(function.is(r1)); // Required by InvokeFunction.
DCHECK(ToRegister(instr->result()).is(r0));
// Copy the arguments to this function possibly from the
// adaptor frame below it.
const uint32_t kArgumentsLimit = 1 * KB;
__ cmp(length, Operand(kArgumentsLimit));
DeoptimizeIf(hi, instr, Deoptimizer::kTooManyArguments);
// Push the receiver and use the register to keep the original
// number of arguments.
__ push(receiver);
__ mov(receiver, length);
// The arguments are at a one pointer size offset from elements.
__ add(elements, elements, Operand(1 * kPointerSize));
// Loop through the arguments pushing them onto the execution
// stack.
Label invoke, loop;
// length is a small non-negative integer, due to the test above.
__ cmp(length, Operand::Zero());
__ b(eq, &invoke);
__ bind(&loop);
__ ldr(scratch, MemOperand(elements, length, LSL, 2));
__ push(scratch);
__ sub(length, length, Operand(1), SetCC);
__ b(ne, &loop);
__ bind(&invoke);
DCHECK(instr->HasPointerMap());
LPointerMap* pointers = instr->pointer_map();
SafepointGenerator safepoint_generator(
this, pointers, Safepoint::kLazyDeopt);
// The number of arguments is stored in receiver which is r0, as expected
// by InvokeFunction.
ParameterCount actual(receiver);
__ InvokeFunction(function, actual, CALL_FUNCTION, safepoint_generator);
}
void LCodeGen::DoPushArgument(LPushArgument* instr) {
LOperand* argument = instr->value();
if (argument->IsDoubleRegister() || argument->IsDoubleStackSlot()) {
Abort(kDoPushArgumentNotImplementedForDoubleType);
} else {
Register argument_reg = EmitLoadRegister(argument, ip);
__ push(argument_reg);
}
}
void LCodeGen::DoDrop(LDrop* instr) {
__ Drop(instr->count());
}
void LCodeGen::DoThisFunction(LThisFunction* instr) {
Register result = ToRegister(instr->result());
__ ldr(result, MemOperand(fp, JavaScriptFrameConstants::kFunctionOffset));
}
void LCodeGen::DoContext(LContext* instr) {
// If there is a non-return use, the context must be moved to a register.
Register result = ToRegister(instr->result());
if (info()->IsOptimizing()) {
__ ldr(result, MemOperand(fp, StandardFrameConstants::kContextOffset));
} else {
// If there is no frame, the context must be in cp.
DCHECK(result.is(cp));
}
}
void LCodeGen::DoDeclareGlobals(LDeclareGlobals* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
__ push(cp); // The context is the first argument.
__ Move(scratch0(), instr->hydrogen()->pairs());
__ push(scratch0());
__ mov(scratch0(), Operand(Smi::FromInt(instr->hydrogen()->flags())));
__ push(scratch0());
CallRuntime(Runtime::kDeclareGlobals, 3, instr);
}
void LCodeGen::CallKnownFunction(Handle<JSFunction> function,
int formal_parameter_count, int arity,
LInstruction* instr) {
bool dont_adapt_arguments =
formal_parameter_count == SharedFunctionInfo::kDontAdaptArgumentsSentinel;
bool can_invoke_directly =
dont_adapt_arguments || formal_parameter_count == arity;
Register function_reg = r1;
LPointerMap* pointers = instr->pointer_map();
if (can_invoke_directly) {
// Change context.
__ ldr(cp, FieldMemOperand(function_reg, JSFunction::kContextOffset));
// Set r0 to arguments count if adaption is not needed. Assumes that r0
// is available to write to at this point.
if (dont_adapt_arguments) {
__ mov(r0, Operand(arity));
}
// Invoke function.
__ ldr(ip, FieldMemOperand(function_reg, JSFunction::kCodeEntryOffset));
__ Call(ip);
// Set up deoptimization.
RecordSafepointWithLazyDeopt(instr, RECORD_SIMPLE_SAFEPOINT);
} else {
SafepointGenerator generator(this, pointers, Safepoint::kLazyDeopt);
ParameterCount count(arity);
ParameterCount expected(formal_parameter_count);
__ InvokeFunction(function_reg, expected, count, CALL_FUNCTION, generator);
}
}
void LCodeGen::DoDeferredMathAbsTaggedHeapNumber(LMathAbs* instr) {
DCHECK(instr->context() != NULL);
DCHECK(ToRegister(instr->context()).is(cp));
Register input = ToRegister(instr->value());
Register result = ToRegister(instr->result());
Register scratch = scratch0();
// Deoptimize if not a heap number.
__ ldr(scratch, FieldMemOperand(input, HeapObject::kMapOffset));
__ LoadRoot(ip, Heap::kHeapNumberMapRootIndex);
__ cmp(scratch, Operand(ip));
DeoptimizeIf(ne, instr, Deoptimizer::kNotAHeapNumber);
Label done;
Register exponent = scratch0();
scratch = no_reg;
__ ldr(exponent, FieldMemOperand(input, HeapNumber::kExponentOffset));
// Check the sign of the argument. If the argument is positive, just
// return it.
__ tst(exponent, Operand(HeapNumber::kSignMask));
// Move the input to the result if necessary.
__ Move(result, input);
__ b(eq, &done);
// Input is negative. Reverse its sign.
// Preserve the value of all registers.
{
PushSafepointRegistersScope scope(this);
// Registers were saved at the safepoint, so we can use
// many scratch registers.
Register tmp1 = input.is(r1) ? r0 : r1;
Register tmp2 = input.is(r2) ? r0 : r2;
Register tmp3 = input.is(r3) ? r0 : r3;
Register tmp4 = input.is(r4) ? r0 : r4;
// exponent: floating point exponent value.
Label allocated, slow;
__ LoadRoot(tmp4, Heap::kHeapNumberMapRootIndex);
__ AllocateHeapNumber(tmp1, tmp2, tmp3, tmp4, &slow);
__ b(&allocated);
// Slow case: Call the runtime system to do the number allocation.
__ bind(&slow);
CallRuntimeFromDeferred(Runtime::kAllocateHeapNumber, 0, instr,
instr->context());
// Set the pointer to the new heap number in tmp.
if (!tmp1.is(r0)) __ mov(tmp1, Operand(r0));
// Restore input_reg after call to runtime.
__ LoadFromSafepointRegisterSlot(input, input);
__ ldr(exponent, FieldMemOperand(input, HeapNumber::kExponentOffset));
__ bind(&allocated);
// exponent: floating point exponent value.
// tmp1: allocated heap number.
__ bic(exponent, exponent, Operand(HeapNumber::kSignMask));
__ str(exponent, FieldMemOperand(tmp1, HeapNumber::kExponentOffset));
__ ldr(tmp2, FieldMemOperand(input, HeapNumber::kMantissaOffset));
__ str(tmp2, FieldMemOperand(tmp1, HeapNumber::kMantissaOffset));
__ StoreToSafepointRegisterSlot(tmp1, result);
}
__ bind(&done);
}
void LCodeGen::EmitIntegerMathAbs(LMathAbs* instr) {
Register input = ToRegister(instr->value());
Register result = ToRegister(instr->result());
__ cmp(input, Operand::Zero());
__ Move(result, input, pl);
// We can make rsb conditional because the previous cmp instruction
// will clear the V (overflow) flag and rsb won't set this flag
// if input is positive.
__ rsb(result, input, Operand::Zero(), SetCC, mi);
// Deoptimize on overflow.
DeoptimizeIf(vs, instr, Deoptimizer::kOverflow);
}
void LCodeGen::DoMathAbs(LMathAbs* instr) {
// Class for deferred case.
class DeferredMathAbsTaggedHeapNumber FINAL : public LDeferredCode {
public:
DeferredMathAbsTaggedHeapNumber(LCodeGen* codegen, LMathAbs* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE {
codegen()->DoDeferredMathAbsTaggedHeapNumber(instr_);
}
LInstruction* instr() OVERRIDE { return instr_; }
private:
LMathAbs* instr_;
};
Representation r = instr->hydrogen()->value()->representation();
if (r.IsDouble()) {
DwVfpRegister input = ToDoubleRegister(instr->value());
DwVfpRegister result = ToDoubleRegister(instr->result());
__ vabs(result, input);
} else if (r.IsSmiOrInteger32()) {
EmitIntegerMathAbs(instr);
} else {
// Representation is tagged.
DeferredMathAbsTaggedHeapNumber* deferred =
new(zone()) DeferredMathAbsTaggedHeapNumber(this, instr);
Register input = ToRegister(instr->value());
// Smi check.
__ JumpIfNotSmi(input, deferred->entry());
// If smi, handle it directly.
EmitIntegerMathAbs(instr);
__ bind(deferred->exit());
}
}
void LCodeGen::DoMathFloor(LMathFloor* instr) {
DwVfpRegister input = ToDoubleRegister(instr->value());
Register result = ToRegister(instr->result());
Register input_high = scratch0();
Label done, exact;
__ TryInt32Floor(result, input, input_high, double_scratch0(), &done, &exact);
DeoptimizeIf(al, instr, Deoptimizer::kLostPrecisionOrNaN);
__ bind(&exact);
if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
// Test for -0.
__ cmp(result, Operand::Zero());
__ b(ne, &done);
__ cmp(input_high, Operand::Zero());
DeoptimizeIf(mi, instr, Deoptimizer::kMinusZero);
}
__ bind(&done);
}
void LCodeGen::DoMathRound(LMathRound* instr) {
DwVfpRegister input = ToDoubleRegister(instr->value());
Register result = ToRegister(instr->result());
DwVfpRegister double_scratch1 = ToDoubleRegister(instr->temp());
DwVfpRegister input_plus_dot_five = double_scratch1;
Register input_high = scratch0();
DwVfpRegister dot_five = double_scratch0();
Label convert, done;
__ Vmov(dot_five, 0.5, scratch0());
__ vabs(double_scratch1, input);
__ VFPCompareAndSetFlags(double_scratch1, dot_five);
// If input is in [-0.5, -0], the result is -0.
// If input is in [+0, +0.5[, the result is +0.
// If the input is +0.5, the result is 1.
__ b(hi, &convert); // Out of [-0.5, +0.5].
if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
__ VmovHigh(input_high, input);
__ cmp(input_high, Operand::Zero());
// [-0.5, -0].
DeoptimizeIf(mi, instr, Deoptimizer::kMinusZero);
}
__ VFPCompareAndSetFlags(input, dot_five);
__ mov(result, Operand(1), LeaveCC, eq); // +0.5.
// Remaining cases: [+0, +0.5[ or [-0.5, +0.5[, depending on
// flag kBailoutOnMinusZero.
__ mov(result, Operand::Zero(), LeaveCC, ne);
__ b(&done);
__ bind(&convert);
__ vadd(input_plus_dot_five, input, dot_five);
// Reuse dot_five (double_scratch0) as we no longer need this value.
__ TryInt32Floor(result, input_plus_dot_five, input_high, double_scratch0(),
&done, &done);
DeoptimizeIf(al, instr, Deoptimizer::kLostPrecisionOrNaN);
__ bind(&done);
}
void LCodeGen::DoMathFround(LMathFround* instr) {
DwVfpRegister input_reg = ToDoubleRegister(instr->value());
DwVfpRegister output_reg = ToDoubleRegister(instr->result());
LowDwVfpRegister scratch = double_scratch0();
__ vcvt_f32_f64(scratch.low(), input_reg);
__ vcvt_f64_f32(output_reg, scratch.low());
}
void LCodeGen::DoMathSqrt(LMathSqrt* instr) {
DwVfpRegister input = ToDoubleRegister(instr->value());
DwVfpRegister result = ToDoubleRegister(instr->result());
__ vsqrt(result, input);
}
void LCodeGen::DoMathPowHalf(LMathPowHalf* instr) {
DwVfpRegister input = ToDoubleRegister(instr->value());
DwVfpRegister result = ToDoubleRegister(instr->result());
DwVfpRegister temp = double_scratch0();
// Note that according to ECMA-262 15.8.2.13:
// Math.pow(-Infinity, 0.5) == Infinity
// Math.sqrt(-Infinity) == NaN
Label done;
__ vmov(temp, -V8_INFINITY, scratch0());
__ VFPCompareAndSetFlags(input, temp);
__ vneg(result, temp, eq);
__ b(&done, eq);
// Add +0 to convert -0 to +0.
__ vadd(result, input, kDoubleRegZero);
__ vsqrt(result, result);
__ bind(&done);
}
void LCodeGen::DoPower(LPower* instr) {
Representation exponent_type = instr->hydrogen()->right()->representation();
// Having marked this as a call, we can use any registers.
// Just make sure that the input/output registers are the expected ones.
Register tagged_exponent = MathPowTaggedDescriptor::exponent();
DCHECK(!instr->right()->IsDoubleRegister() ||
ToDoubleRegister(instr->right()).is(d1));
DCHECK(!instr->right()->IsRegister() ||
ToRegister(instr->right()).is(tagged_exponent));
DCHECK(ToDoubleRegister(instr->left()).is(d0));
DCHECK(ToDoubleRegister(instr->result()).is(d2));
if (exponent_type.IsSmi()) {
MathPowStub stub(isolate(), MathPowStub::TAGGED);
__ CallStub(&stub);
} else if (exponent_type.IsTagged()) {
Label no_deopt;
__ JumpIfSmi(tagged_exponent, &no_deopt);
DCHECK(!r6.is(tagged_exponent));
__ ldr(r6, FieldMemOperand(tagged_exponent, HeapObject::kMapOffset));
__ LoadRoot(ip, Heap::kHeapNumberMapRootIndex);
__ cmp(r6, Operand(ip));
DeoptimizeIf(ne, instr, Deoptimizer::kNotAHeapNumber);
__ bind(&no_deopt);
MathPowStub stub(isolate(), MathPowStub::TAGGED);
__ CallStub(&stub);
} else if (exponent_type.IsInteger32()) {
MathPowStub stub(isolate(), MathPowStub::INTEGER);
__ CallStub(&stub);
} else {
DCHECK(exponent_type.IsDouble());
MathPowStub stub(isolate(), MathPowStub::DOUBLE);
__ CallStub(&stub);
}
}
void LCodeGen::DoMathExp(LMathExp* instr) {
DwVfpRegister input = ToDoubleRegister(instr->value());
DwVfpRegister result = ToDoubleRegister(instr->result());
DwVfpRegister double_scratch1 = ToDoubleRegister(instr->double_temp());
DwVfpRegister double_scratch2 = double_scratch0();
Register temp1 = ToRegister(instr->temp1());
Register temp2 = ToRegister(instr->temp2());
MathExpGenerator::EmitMathExp(
masm(), input, result, double_scratch1, double_scratch2,
temp1, temp2, scratch0());
}
void LCodeGen::DoMathLog(LMathLog* instr) {
__ PrepareCallCFunction(0, 1, scratch0());
__ MovToFloatParameter(ToDoubleRegister(instr->value()));
__ CallCFunction(ExternalReference::math_log_double_function(isolate()),
0, 1);
__ MovFromFloatResult(ToDoubleRegister(instr->result()));
}
void LCodeGen::DoMathClz32(LMathClz32* instr) {
Register input = ToRegister(instr->value());
Register result = ToRegister(instr->result());
__ clz(result, input);
}
void LCodeGen::DoInvokeFunction(LInvokeFunction* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->function()).is(r1));
DCHECK(instr->HasPointerMap());
Handle<JSFunction> known_function = instr->hydrogen()->known_function();
if (known_function.is_null()) {
LPointerMap* pointers = instr->pointer_map();
SafepointGenerator generator(this, pointers, Safepoint::kLazyDeopt);
ParameterCount count(instr->arity());
__ InvokeFunction(r1, count, CALL_FUNCTION, generator);
} else {
CallKnownFunction(known_function,
instr->hydrogen()->formal_parameter_count(),
instr->arity(), instr);
}
}
void LCodeGen::DoTailCallThroughMegamorphicCache(
LTailCallThroughMegamorphicCache* instr) {
Register receiver = ToRegister(instr->receiver());
Register name = ToRegister(instr->name());
DCHECK(receiver.is(LoadDescriptor::ReceiverRegister()));
DCHECK(name.is(LoadDescriptor::NameRegister()));
DCHECK(receiver.is(r1));
DCHECK(name.is(r2));
Register scratch = r4;
Register extra = r5;
Register extra2 = r6;
Register extra3 = r9;
#ifdef DEBUG
Register slot = FLAG_vector_ics ? ToRegister(instr->slot()) : no_reg;
Register vector = FLAG_vector_ics ? ToRegister(instr->vector()) : no_reg;
DCHECK(!FLAG_vector_ics ||
!AreAliased(slot, vector, scratch, extra, extra2, extra3));
#endif
// Important for the tail-call.
bool must_teardown_frame = NeedsEagerFrame();
if (!instr->hydrogen()->is_just_miss()) {
DCHECK(!instr->hydrogen()->is_keyed_load());
// The probe will tail call to a handler if found.
isolate()->stub_cache()->GenerateProbe(
masm(), Code::LOAD_IC, instr->hydrogen()->flags(), must_teardown_frame,
receiver, name, scratch, extra, extra2, extra3);
}
// Tail call to miss if we ended up here.
if (must_teardown_frame) __ LeaveFrame(StackFrame::INTERNAL);
if (instr->hydrogen()->is_keyed_load()) {
KeyedLoadIC::GenerateMiss(masm());
} else {
LoadIC::GenerateMiss(masm());
}
}
This is a preview of a first step towards unification of the hydrogen call machinery. The change replaces CallNamed, CallKeyed, CallConstantFunction and CallKnownGlobal hydrogen instructions with two new instructions with a more lower level semantics: 1. CallJSFunction for direct calls of JSFunction objects (no argument adaptation) 2. CallWithDescriptor for calls of a given Code object according to the supplied calling convention. Details: CallJSFunction should be straightforward, the main difference from the existing InvokeFunction instruction is the absence of argument adaptor handling. (As a next step, we will replace InvokeFunction with an equivalent hydrogen code.) For CallWithDescriptor, the calling conventions are represented by a tweaked version of CallStubInterfaceDescriptor. In addition to the parameter-register mapping, we also define parameter-representation mapping there. The CallWithDescriptor instruction has variable number of parameters now - this required some simple tweaks in Lithium, which assumed fixed number of arguments in some places. The calling conventions used in the calls are initialized in the CallDescriptors class (code-stubs.h, <arch>/code-stubs-<arch>.cc), and they live in a new table in the Isolate class. I should say I am not quite sure about Representation::Integer32() representation for some of the params of ArgumentAdaptorCall - it is not clear to me wether the params could not end up on the stack and thus confuse the GC. The change also includes an earlier small change to argument adaptor (https://codereview.chromium.org/98463007) that avoids passing a naked pointer to the code entry as a parameter. I am sorry for packaging that with an already biggish change. Performance implications: Locally, I see a small regression (.2% or so). It is hard to say where exactly it comes from, but I do see inefficient call sequences to the adaptor trampoline. For example: ;;; <@78,#24> constant-t bf85aa515a mov edi,0x5a51aa85 ;; debug: position 29 ;;; <@72,#53> load-named-field 8b7717 mov esi,[edi+0x17] ;; debug: position 195 ;;; <@80,#51> constant-s b902000000 mov ecx,0x2 ;; debug: position 195 ;;; <@81,#51> gap 894df0 mov [ebp+0xf0],ecx ;;; <@82,#103> constant-i bb01000000 mov ebx,0x1 ;;; <@84,#102> constant-i b902000000 mov ecx,0x2 ;;; <@85,#102> gap 89d8 mov eax,ebx 89cb mov ebx,ecx 8b4df0 mov ecx,[ebp+0xf0] ;;; <@86,#58> call-with-descriptor e8ef57fcff call ArgumentsAdaptorTrampoline (0x2d80e6e0) ;; code: BUILTIN Note the silly handling of ecx; the hydrogen for this code is: 0 4 s27 Constant 1 range:1_1 <|@ 0 3 t30 Constant 0x5bc1aa85 <JS Function xyz (SharedFunctionInfo 0x5bc1a919)> type:object <|@ 0 1 t36 LoadNamedField t30.[in-object]@24 <|@ 0 1 t38 Constant 0x2300e6a1 <Code> <|@ 0 1 i102 Constant 2 range:2_2 <|@ 0 1 i103 Constant 1 range:1_1 <|@ 0 2 t41 CallWithDescriptor t38 t30 t36 s27 i103 i102 #2 changes[*] <|@ BUG= R=verwaest@chromium.org, danno@chromium.org Review URL: https://codereview.chromium.org/104663004 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@18626 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-01-15 17:00:35 +00:00
void LCodeGen::DoCallWithDescriptor(LCallWithDescriptor* instr) {
DCHECK(ToRegister(instr->result()).is(r0));
if (instr->hydrogen()->IsTailCall()) {
if (NeedsEagerFrame()) __ LeaveFrame(StackFrame::INTERNAL);
if (instr->target()->IsConstantOperand()) {
LConstantOperand* target = LConstantOperand::cast(instr->target());
Handle<Code> code = Handle<Code>::cast(ToHandle(target));
__ Jump(code, RelocInfo::CODE_TARGET);
} else {
DCHECK(instr->target()->IsRegister());
Register target = ToRegister(instr->target());
// Make sure we don't emit any additional entries in the constant pool
// before the call to ensure that the CallCodeSize() calculated the
// correct
// number of instructions for the constant pool load.
{
ConstantPoolUnavailableScope constant_pool_unavailable(masm_);
__ add(target, target, Operand(Code::kHeaderSize - kHeapObjectTag));
}
__ Jump(target);
}
This is a preview of a first step towards unification of the hydrogen call machinery. The change replaces CallNamed, CallKeyed, CallConstantFunction and CallKnownGlobal hydrogen instructions with two new instructions with a more lower level semantics: 1. CallJSFunction for direct calls of JSFunction objects (no argument adaptation) 2. CallWithDescriptor for calls of a given Code object according to the supplied calling convention. Details: CallJSFunction should be straightforward, the main difference from the existing InvokeFunction instruction is the absence of argument adaptor handling. (As a next step, we will replace InvokeFunction with an equivalent hydrogen code.) For CallWithDescriptor, the calling conventions are represented by a tweaked version of CallStubInterfaceDescriptor. In addition to the parameter-register mapping, we also define parameter-representation mapping there. The CallWithDescriptor instruction has variable number of parameters now - this required some simple tweaks in Lithium, which assumed fixed number of arguments in some places. The calling conventions used in the calls are initialized in the CallDescriptors class (code-stubs.h, <arch>/code-stubs-<arch>.cc), and they live in a new table in the Isolate class. I should say I am not quite sure about Representation::Integer32() representation for some of the params of ArgumentAdaptorCall - it is not clear to me wether the params could not end up on the stack and thus confuse the GC. The change also includes an earlier small change to argument adaptor (https://codereview.chromium.org/98463007) that avoids passing a naked pointer to the code entry as a parameter. I am sorry for packaging that with an already biggish change. Performance implications: Locally, I see a small regression (.2% or so). It is hard to say where exactly it comes from, but I do see inefficient call sequences to the adaptor trampoline. For example: ;;; <@78,#24> constant-t bf85aa515a mov edi,0x5a51aa85 ;; debug: position 29 ;;; <@72,#53> load-named-field 8b7717 mov esi,[edi+0x17] ;; debug: position 195 ;;; <@80,#51> constant-s b902000000 mov ecx,0x2 ;; debug: position 195 ;;; <@81,#51> gap 894df0 mov [ebp+0xf0],ecx ;;; <@82,#103> constant-i bb01000000 mov ebx,0x1 ;;; <@84,#102> constant-i b902000000 mov ecx,0x2 ;;; <@85,#102> gap 89d8 mov eax,ebx 89cb mov ebx,ecx 8b4df0 mov ecx,[ebp+0xf0] ;;; <@86,#58> call-with-descriptor e8ef57fcff call ArgumentsAdaptorTrampoline (0x2d80e6e0) ;; code: BUILTIN Note the silly handling of ecx; the hydrogen for this code is: 0 4 s27 Constant 1 range:1_1 <|@ 0 3 t30 Constant 0x5bc1aa85 <JS Function xyz (SharedFunctionInfo 0x5bc1a919)> type:object <|@ 0 1 t36 LoadNamedField t30.[in-object]@24 <|@ 0 1 t38 Constant 0x2300e6a1 <Code> <|@ 0 1 i102 Constant 2 range:2_2 <|@ 0 1 i103 Constant 1 range:1_1 <|@ 0 2 t41 CallWithDescriptor t38 t30 t36 s27 i103 i102 #2 changes[*] <|@ BUG= R=verwaest@chromium.org, danno@chromium.org Review URL: https://codereview.chromium.org/104663004 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@18626 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-01-15 17:00:35 +00:00
} else {
LPointerMap* pointers = instr->pointer_map();
SafepointGenerator generator(this, pointers, Safepoint::kLazyDeopt);
if (instr->target()->IsConstantOperand()) {
LConstantOperand* target = LConstantOperand::cast(instr->target());
Handle<Code> code = Handle<Code>::cast(ToHandle(target));
generator.BeforeCall(__ CallSize(code, RelocInfo::CODE_TARGET));
PlatformInterfaceDescriptor* call_descriptor =
instr->descriptor().platform_specific_descriptor();
__ Call(code, RelocInfo::CODE_TARGET, TypeFeedbackId::None(), al,
call_descriptor->storage_mode());
} else {
DCHECK(instr->target()->IsRegister());
Register target = ToRegister(instr->target());
generator.BeforeCall(__ CallSize(target));
// Make sure we don't emit any additional entries in the constant pool
// before the call to ensure that the CallCodeSize() calculated the
// correct
// number of instructions for the constant pool load.
{
ConstantPoolUnavailableScope constant_pool_unavailable(masm_);
__ add(target, target, Operand(Code::kHeaderSize - kHeapObjectTag));
}
__ Call(target);
}
generator.AfterCall();
}
}
This is a preview of a first step towards unification of the hydrogen call machinery. The change replaces CallNamed, CallKeyed, CallConstantFunction and CallKnownGlobal hydrogen instructions with two new instructions with a more lower level semantics: 1. CallJSFunction for direct calls of JSFunction objects (no argument adaptation) 2. CallWithDescriptor for calls of a given Code object according to the supplied calling convention. Details: CallJSFunction should be straightforward, the main difference from the existing InvokeFunction instruction is the absence of argument adaptor handling. (As a next step, we will replace InvokeFunction with an equivalent hydrogen code.) For CallWithDescriptor, the calling conventions are represented by a tweaked version of CallStubInterfaceDescriptor. In addition to the parameter-register mapping, we also define parameter-representation mapping there. The CallWithDescriptor instruction has variable number of parameters now - this required some simple tweaks in Lithium, which assumed fixed number of arguments in some places. The calling conventions used in the calls are initialized in the CallDescriptors class (code-stubs.h, <arch>/code-stubs-<arch>.cc), and they live in a new table in the Isolate class. I should say I am not quite sure about Representation::Integer32() representation for some of the params of ArgumentAdaptorCall - it is not clear to me wether the params could not end up on the stack and thus confuse the GC. The change also includes an earlier small change to argument adaptor (https://codereview.chromium.org/98463007) that avoids passing a naked pointer to the code entry as a parameter. I am sorry for packaging that with an already biggish change. Performance implications: Locally, I see a small regression (.2% or so). It is hard to say where exactly it comes from, but I do see inefficient call sequences to the adaptor trampoline. For example: ;;; <@78,#24> constant-t bf85aa515a mov edi,0x5a51aa85 ;; debug: position 29 ;;; <@72,#53> load-named-field 8b7717 mov esi,[edi+0x17] ;; debug: position 195 ;;; <@80,#51> constant-s b902000000 mov ecx,0x2 ;; debug: position 195 ;;; <@81,#51> gap 894df0 mov [ebp+0xf0],ecx ;;; <@82,#103> constant-i bb01000000 mov ebx,0x1 ;;; <@84,#102> constant-i b902000000 mov ecx,0x2 ;;; <@85,#102> gap 89d8 mov eax,ebx 89cb mov ebx,ecx 8b4df0 mov ecx,[ebp+0xf0] ;;; <@86,#58> call-with-descriptor e8ef57fcff call ArgumentsAdaptorTrampoline (0x2d80e6e0) ;; code: BUILTIN Note the silly handling of ecx; the hydrogen for this code is: 0 4 s27 Constant 1 range:1_1 <|@ 0 3 t30 Constant 0x5bc1aa85 <JS Function xyz (SharedFunctionInfo 0x5bc1a919)> type:object <|@ 0 1 t36 LoadNamedField t30.[in-object]@24 <|@ 0 1 t38 Constant 0x2300e6a1 <Code> <|@ 0 1 i102 Constant 2 range:2_2 <|@ 0 1 i103 Constant 1 range:1_1 <|@ 0 2 t41 CallWithDescriptor t38 t30 t36 s27 i103 i102 #2 changes[*] <|@ BUG= R=verwaest@chromium.org, danno@chromium.org Review URL: https://codereview.chromium.org/104663004 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@18626 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-01-15 17:00:35 +00:00
void LCodeGen::DoCallJSFunction(LCallJSFunction* instr) {
DCHECK(ToRegister(instr->function()).is(r1));
DCHECK(ToRegister(instr->result()).is(r0));
This is a preview of a first step towards unification of the hydrogen call machinery. The change replaces CallNamed, CallKeyed, CallConstantFunction and CallKnownGlobal hydrogen instructions with two new instructions with a more lower level semantics: 1. CallJSFunction for direct calls of JSFunction objects (no argument adaptation) 2. CallWithDescriptor for calls of a given Code object according to the supplied calling convention. Details: CallJSFunction should be straightforward, the main difference from the existing InvokeFunction instruction is the absence of argument adaptor handling. (As a next step, we will replace InvokeFunction with an equivalent hydrogen code.) For CallWithDescriptor, the calling conventions are represented by a tweaked version of CallStubInterfaceDescriptor. In addition to the parameter-register mapping, we also define parameter-representation mapping there. The CallWithDescriptor instruction has variable number of parameters now - this required some simple tweaks in Lithium, which assumed fixed number of arguments in some places. The calling conventions used in the calls are initialized in the CallDescriptors class (code-stubs.h, <arch>/code-stubs-<arch>.cc), and they live in a new table in the Isolate class. I should say I am not quite sure about Representation::Integer32() representation for some of the params of ArgumentAdaptorCall - it is not clear to me wether the params could not end up on the stack and thus confuse the GC. The change also includes an earlier small change to argument adaptor (https://codereview.chromium.org/98463007) that avoids passing a naked pointer to the code entry as a parameter. I am sorry for packaging that with an already biggish change. Performance implications: Locally, I see a small regression (.2% or so). It is hard to say where exactly it comes from, but I do see inefficient call sequences to the adaptor trampoline. For example: ;;; <@78,#24> constant-t bf85aa515a mov edi,0x5a51aa85 ;; debug: position 29 ;;; <@72,#53> load-named-field 8b7717 mov esi,[edi+0x17] ;; debug: position 195 ;;; <@80,#51> constant-s b902000000 mov ecx,0x2 ;; debug: position 195 ;;; <@81,#51> gap 894df0 mov [ebp+0xf0],ecx ;;; <@82,#103> constant-i bb01000000 mov ebx,0x1 ;;; <@84,#102> constant-i b902000000 mov ecx,0x2 ;;; <@85,#102> gap 89d8 mov eax,ebx 89cb mov ebx,ecx 8b4df0 mov ecx,[ebp+0xf0] ;;; <@86,#58> call-with-descriptor e8ef57fcff call ArgumentsAdaptorTrampoline (0x2d80e6e0) ;; code: BUILTIN Note the silly handling of ecx; the hydrogen for this code is: 0 4 s27 Constant 1 range:1_1 <|@ 0 3 t30 Constant 0x5bc1aa85 <JS Function xyz (SharedFunctionInfo 0x5bc1a919)> type:object <|@ 0 1 t36 LoadNamedField t30.[in-object]@24 <|@ 0 1 t38 Constant 0x2300e6a1 <Code> <|@ 0 1 i102 Constant 2 range:2_2 <|@ 0 1 i103 Constant 1 range:1_1 <|@ 0 2 t41 CallWithDescriptor t38 t30 t36 s27 i103 i102 #2 changes[*] <|@ BUG= R=verwaest@chromium.org, danno@chromium.org Review URL: https://codereview.chromium.org/104663004 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@18626 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2014-01-15 17:00:35 +00:00
if (instr->hydrogen()->pass_argument_count()) {
__ mov(r0, Operand(instr->arity()));
}
// Change context.
__ ldr(cp, FieldMemOperand(r1, JSFunction::kContextOffset));
// Load the code entry address
__ ldr(ip, FieldMemOperand(r1, JSFunction::kCodeEntryOffset));
__ Call(ip);
RecordSafepointWithLazyDeopt(instr, RECORD_SIMPLE_SAFEPOINT);
}
void LCodeGen::DoCallFunction(LCallFunction* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->function()).is(r1));
DCHECK(ToRegister(instr->result()).is(r0));
int arity = instr->arity();
CallFunctionFlags flags = instr->hydrogen()->function_flags();
if (instr->hydrogen()->HasVectorAndSlot()) {
Register slot_register = ToRegister(instr->temp_slot());
Register vector_register = ToRegister(instr->temp_vector());
DCHECK(slot_register.is(r3));
DCHECK(vector_register.is(r2));
AllowDeferredHandleDereference vector_structure_check;
Handle<TypeFeedbackVector> vector = instr->hydrogen()->feedback_vector();
int index = vector->GetIndex(instr->hydrogen()->slot());
__ Move(vector_register, vector);
__ mov(slot_register, Operand(Smi::FromInt(index)));
CallICState::CallType call_type =
(flags & CALL_AS_METHOD) ? CallICState::METHOD : CallICState::FUNCTION;
Handle<Code> ic =
CodeFactory::CallICInOptimizedCode(isolate(), arity, call_type).code();
CallCode(ic, RelocInfo::CODE_TARGET, instr);
} else {
CallFunctionStub stub(isolate(), arity, flags);
CallCode(stub.GetCode(), RelocInfo::CODE_TARGET, instr);
}
}
void LCodeGen::DoCallNew(LCallNew* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->constructor()).is(r1));
DCHECK(ToRegister(instr->result()).is(r0));
__ mov(r0, Operand(instr->arity()));
// No cell in r2 for construct type feedback in optimized code
__ LoadRoot(r2, Heap::kUndefinedValueRootIndex);
CallConstructStub stub(isolate(), NO_CALL_CONSTRUCTOR_FLAGS);
CallCode(stub.GetCode(), RelocInfo::CONSTRUCT_CALL, instr);
}
void LCodeGen::DoCallNewArray(LCallNewArray* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->constructor()).is(r1));
DCHECK(ToRegister(instr->result()).is(r0));
__ mov(r0, Operand(instr->arity()));
__ LoadRoot(r2, Heap::kUndefinedValueRootIndex);
ElementsKind kind = instr->hydrogen()->elements_kind();
AllocationSiteOverrideMode override_mode =
(AllocationSite::GetMode(kind) == TRACK_ALLOCATION_SITE)
? DISABLE_ALLOCATION_SITES
: DONT_OVERRIDE;
if (instr->arity() == 0) {
ArrayNoArgumentConstructorStub stub(isolate(), kind, override_mode);
CallCode(stub.GetCode(), RelocInfo::CONSTRUCT_CALL, instr);
} else if (instr->arity() == 1) {
Label done;
if (IsFastPackedElementsKind(kind)) {
Label packed_case;
// We might need a change here
// look at the first argument
__ ldr(r5, MemOperand(sp, 0));
__ cmp(r5, Operand::Zero());
__ b(eq, &packed_case);
ElementsKind holey_kind = GetHoleyElementsKind(kind);
ArraySingleArgumentConstructorStub stub(isolate(),
holey_kind,
override_mode);
CallCode(stub.GetCode(), RelocInfo::CONSTRUCT_CALL, instr);
__ jmp(&done);
__ bind(&packed_case);
}
ArraySingleArgumentConstructorStub stub(isolate(), kind, override_mode);
CallCode(stub.GetCode(), RelocInfo::CONSTRUCT_CALL, instr);
__ bind(&done);
} else {
ArrayNArgumentsConstructorStub stub(isolate(), kind, override_mode);
CallCode(stub.GetCode(), RelocInfo::CONSTRUCT_CALL, instr);
}
}
void LCodeGen::DoCallRuntime(LCallRuntime* instr) {
CallRuntime(instr->function(), instr->arity(), instr);
}
void LCodeGen::DoStoreCodeEntry(LStoreCodeEntry* instr) {
Register function = ToRegister(instr->function());
Register code_object = ToRegister(instr->code_object());
__ add(code_object, code_object, Operand(Code::kHeaderSize - kHeapObjectTag));
__ str(code_object,
FieldMemOperand(function, JSFunction::kCodeEntryOffset));
}
void LCodeGen::DoInnerAllocatedObject(LInnerAllocatedObject* instr) {
Register result = ToRegister(instr->result());
Register base = ToRegister(instr->base_object());
if (instr->offset()->IsConstantOperand()) {
LConstantOperand* offset = LConstantOperand::cast(instr->offset());
__ add(result, base, Operand(ToInteger32(offset)));
} else {
Register offset = ToRegister(instr->offset());
__ add(result, base, offset);
}
}
void LCodeGen::DoStoreNamedField(LStoreNamedField* instr) {
Representation representation = instr->representation();
Register object = ToRegister(instr->object());
Register scratch = scratch0();
HObjectAccess access = instr->hydrogen()->access();
int offset = access.offset();
if (access.IsExternalMemory()) {
Register value = ToRegister(instr->value());
MemOperand operand = MemOperand(object, offset);
__ Store(value, operand, representation);
return;
}
__ AssertNotSmi(object);
DCHECK(!representation.IsSmi() ||
!instr->value()->IsConstantOperand() ||
IsSmi(LConstantOperand::cast(instr->value())));
if (representation.IsDouble()) {
DCHECK(access.IsInobject());
DCHECK(!instr->hydrogen()->has_transition());
DCHECK(!instr->hydrogen()->NeedsWriteBarrier());
DwVfpRegister value = ToDoubleRegister(instr->value());
__ vstr(value, FieldMemOperand(object, offset));
return;
}
if (instr->hydrogen()->has_transition()) {
Handle<Map> transition = instr->hydrogen()->transition_map();
AddDeprecationDependency(transition);
__ mov(scratch, Operand(transition));
__ str(scratch, FieldMemOperand(object, HeapObject::kMapOffset));
if (instr->hydrogen()->NeedsWriteBarrierForMap()) {
Register temp = ToRegister(instr->temp());
// Update the write barrier for the map field.
__ RecordWriteForMap(object,
scratch,
temp,
GetLinkRegisterState(),
kSaveFPRegs);
}
}
// Do the store.
Register value = ToRegister(instr->value());
if (access.IsInobject()) {
MemOperand operand = FieldMemOperand(object, offset);
__ Store(value, operand, representation);
if (instr->hydrogen()->NeedsWriteBarrier()) {
// Update the write barrier for the object for in-object properties.
__ RecordWriteField(object,
offset,
value,
scratch,
GetLinkRegisterState(),
kSaveFPRegs,
EMIT_REMEMBERED_SET,
instr->hydrogen()->SmiCheckForWriteBarrier(),
instr->hydrogen()->PointersToHereCheckForValue());
}
} else {
__ ldr(scratch, FieldMemOperand(object, JSObject::kPropertiesOffset));
MemOperand operand = FieldMemOperand(scratch, offset);
__ Store(value, operand, representation);
if (instr->hydrogen()->NeedsWriteBarrier()) {
// Update the write barrier for the properties array.
// object is used as a scratch register.
__ RecordWriteField(scratch,
offset,
value,
object,
GetLinkRegisterState(),
kSaveFPRegs,
EMIT_REMEMBERED_SET,
instr->hydrogen()->SmiCheckForWriteBarrier(),
instr->hydrogen()->PointersToHereCheckForValue());
}
}
}
void LCodeGen::DoStoreNamedGeneric(LStoreNamedGeneric* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->object()).is(StoreDescriptor::ReceiverRegister()));
DCHECK(ToRegister(instr->value()).is(StoreDescriptor::ValueRegister()));
__ mov(StoreDescriptor::NameRegister(), Operand(instr->name()));
Handle<Code> ic =
StoreIC::initialize_stub(isolate(), instr->language_mode(),
instr->hydrogen()->initialization_state());
CallCode(ic, RelocInfo::CODE_TARGET, instr, NEVER_INLINE_TARGET_ADDRESS);
}
void LCodeGen::DoBoundsCheck(LBoundsCheck* instr) {
Condition cc = instr->hydrogen()->allow_equality() ? hi : hs;
if (instr->index()->IsConstantOperand()) {
Operand index = ToOperand(instr->index());
Register length = ToRegister(instr->length());
__ cmp(length, index);
cc = CommuteCondition(cc);
} else {
Register index = ToRegister(instr->index());
Operand length = ToOperand(instr->length());
__ cmp(index, length);
}
if (FLAG_debug_code && instr->hydrogen()->skip_check()) {
Label done;
__ b(NegateCondition(cc), &done);
__ stop("eliminated bounds check failed");
__ bind(&done);
} else {
DeoptimizeIf(cc, instr, Deoptimizer::kOutOfBounds);
}
}
void LCodeGen::DoStoreKeyedExternalArray(LStoreKeyed* instr) {
Register external_pointer = ToRegister(instr->elements());
Register key = no_reg;
ElementsKind elements_kind = instr->elements_kind();
bool key_is_constant = instr->key()->IsConstantOperand();
int constant_key = 0;
if (key_is_constant) {
constant_key = ToInteger32(LConstantOperand::cast(instr->key()));
if (constant_key & 0xF0000000) {
Abort(kArrayIndexConstantValueTooBig);
}
} else {
key = ToRegister(instr->key());
}
int element_size_shift = ElementsKindToShiftSize(elements_kind);
int shift_size = (instr->hydrogen()->key()->representation().IsSmi())
? (element_size_shift - kSmiTagSize) : element_size_shift;
int base_offset = instr->base_offset();
if (elements_kind == EXTERNAL_FLOAT32_ELEMENTS ||
elements_kind == FLOAT32_ELEMENTS ||
elements_kind == EXTERNAL_FLOAT64_ELEMENTS ||
elements_kind == FLOAT64_ELEMENTS) {
Register address = scratch0();
DwVfpRegister value(ToDoubleRegister(instr->value()));
if (key_is_constant) {
if (constant_key != 0) {
__ add(address, external_pointer,
Operand(constant_key << element_size_shift));
} else {
address = external_pointer;
}
} else {
__ add(address, external_pointer, Operand(key, LSL, shift_size));
}
if (elements_kind == EXTERNAL_FLOAT32_ELEMENTS ||
elements_kind == FLOAT32_ELEMENTS) {
__ vcvt_f32_f64(double_scratch0().low(), value);
__ vstr(double_scratch0().low(), address, base_offset);
} else { // Storing doubles, not floats.
__ vstr(value, address, base_offset);
}
} else {
Register value(ToRegister(instr->value()));
MemOperand mem_operand = PrepareKeyedOperand(
key, external_pointer, key_is_constant, constant_key,
element_size_shift, shift_size,
base_offset);
switch (elements_kind) {
case EXTERNAL_UINT8_CLAMPED_ELEMENTS:
case EXTERNAL_INT8_ELEMENTS:
case EXTERNAL_UINT8_ELEMENTS:
case UINT8_ELEMENTS:
case UINT8_CLAMPED_ELEMENTS:
case INT8_ELEMENTS:
__ strb(value, mem_operand);
break;
case EXTERNAL_INT16_ELEMENTS:
case EXTERNAL_UINT16_ELEMENTS:
case INT16_ELEMENTS:
case UINT16_ELEMENTS:
__ strh(value, mem_operand);
break;
case EXTERNAL_INT32_ELEMENTS:
case EXTERNAL_UINT32_ELEMENTS:
case INT32_ELEMENTS:
case UINT32_ELEMENTS:
__ str(value, mem_operand);
break;
case FLOAT32_ELEMENTS:
case FLOAT64_ELEMENTS:
case EXTERNAL_FLOAT32_ELEMENTS:
case EXTERNAL_FLOAT64_ELEMENTS:
case FAST_DOUBLE_ELEMENTS:
case FAST_ELEMENTS:
case FAST_SMI_ELEMENTS:
case FAST_HOLEY_DOUBLE_ELEMENTS:
case FAST_HOLEY_ELEMENTS:
case FAST_HOLEY_SMI_ELEMENTS:
case DICTIONARY_ELEMENTS:
case SLOPPY_ARGUMENTS_ELEMENTS:
UNREACHABLE();
break;
}
}
}
void LCodeGen::DoStoreKeyedFixedDoubleArray(LStoreKeyed* instr) {
DwVfpRegister value = ToDoubleRegister(instr->value());
Register elements = ToRegister(instr->elements());
Register scratch = scratch0();
DwVfpRegister double_scratch = double_scratch0();
bool key_is_constant = instr->key()->IsConstantOperand();
int base_offset = instr->base_offset();
// Calculate the effective address of the slot in the array to store the
// double value.
int element_size_shift = ElementsKindToShiftSize(FAST_DOUBLE_ELEMENTS);
if (key_is_constant) {
int constant_key = ToInteger32(LConstantOperand::cast(instr->key()));
if (constant_key & 0xF0000000) {
Abort(kArrayIndexConstantValueTooBig);
}
__ add(scratch, elements,
Operand((constant_key << element_size_shift) + base_offset));
} else {
int shift_size = (instr->hydrogen()->key()->representation().IsSmi())
? (element_size_shift - kSmiTagSize) : element_size_shift;
__ add(scratch, elements, Operand(base_offset));
__ add(scratch, scratch,
Operand(ToRegister(instr->key()), LSL, shift_size));
}
if (instr->NeedsCanonicalization()) {
// Force a canonical NaN.
if (masm()->emit_debug_code()) {
__ vmrs(ip);
__ tst(ip, Operand(kVFPDefaultNaNModeControlBit));
__ Assert(ne, kDefaultNaNModeNotSet);
}
__ VFPCanonicalizeNaN(double_scratch, value);
__ vstr(double_scratch, scratch, 0);
} else {
__ vstr(value, scratch, 0);
}
}
void LCodeGen::DoStoreKeyedFixedArray(LStoreKeyed* instr) {
Register value = ToRegister(instr->value());
Register elements = ToRegister(instr->elements());
Register key = instr->key()->IsRegister() ? ToRegister(instr->key())
: no_reg;
Register scratch = scratch0();
Register store_base = scratch;
int offset = instr->base_offset();
// Do the store.
if (instr->key()->IsConstantOperand()) {
DCHECK(!instr->hydrogen()->NeedsWriteBarrier());
LConstantOperand* const_operand = LConstantOperand::cast(instr->key());
offset += ToInteger32(const_operand) * kPointerSize;
store_base = elements;
} else {
// Even though the HLoadKeyed instruction forces the input
// representation for the key to be an integer, the input gets replaced
// during bound check elimination with the index argument to the bounds
// check, which can be tagged, so that case must be handled here, too.
if (instr->hydrogen()->key()->representation().IsSmi()) {
__ add(scratch, elements, Operand::PointerOffsetFromSmiKey(key));
} else {
__ add(scratch, elements, Operand(key, LSL, kPointerSizeLog2));
}
}
__ str(value, MemOperand(store_base, offset));
if (instr->hydrogen()->NeedsWriteBarrier()) {
SmiCheck check_needed =
instr->hydrogen()->value()->type().IsHeapObject()
? OMIT_SMI_CHECK : INLINE_SMI_CHECK;
// Compute address of modified element and store it into key register.
__ add(key, store_base, Operand(offset));
__ RecordWrite(elements,
key,
value,
GetLinkRegisterState(),
kSaveFPRegs,
EMIT_REMEMBERED_SET,
check_needed,
instr->hydrogen()->PointersToHereCheckForValue());
}
}
void LCodeGen::DoStoreKeyed(LStoreKeyed* instr) {
// By cases: external, fast double
if (instr->is_typed_elements()) {
DoStoreKeyedExternalArray(instr);
} else if (instr->hydrogen()->value()->representation().IsDouble()) {
DoStoreKeyedFixedDoubleArray(instr);
} else {
DoStoreKeyedFixedArray(instr);
}
}
void LCodeGen::DoStoreKeyedGeneric(LStoreKeyedGeneric* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->object()).is(StoreDescriptor::ReceiverRegister()));
DCHECK(ToRegister(instr->key()).is(StoreDescriptor::NameRegister()));
DCHECK(ToRegister(instr->value()).is(StoreDescriptor::ValueRegister()));
Handle<Code> ic = CodeFactory::KeyedStoreICInOptimizedCode(
isolate(), instr->language_mode(),
instr->hydrogen()->initialization_state()).code();
CallCode(ic, RelocInfo::CODE_TARGET, instr, NEVER_INLINE_TARGET_ADDRESS);
}
void LCodeGen::DoTransitionElementsKind(LTransitionElementsKind* instr) {
Register object_reg = ToRegister(instr->object());
Register scratch = scratch0();
Handle<Map> from_map = instr->original_map();
Handle<Map> to_map = instr->transitioned_map();
ElementsKind from_kind = instr->from_kind();
ElementsKind to_kind = instr->to_kind();
Label not_applicable;
__ ldr(scratch, FieldMemOperand(object_reg, HeapObject::kMapOffset));
__ cmp(scratch, Operand(from_map));
__ b(ne, &not_applicable);
if (IsSimpleMapChangeTransition(from_kind, to_kind)) {
Register new_map_reg = ToRegister(instr->new_map_temp());
__ mov(new_map_reg, Operand(to_map));
__ str(new_map_reg, FieldMemOperand(object_reg, HeapObject::kMapOffset));
// Write barrier.
__ RecordWriteForMap(object_reg,
new_map_reg,
scratch,
GetLinkRegisterState(),
kDontSaveFPRegs);
} else {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(object_reg.is(r0));
PushSafepointRegistersScope scope(this);
__ Move(r1, to_map);
bool is_js_array = from_map->instance_type() == JS_ARRAY_TYPE;
TransitionElementsKindStub stub(isolate(), from_kind, to_kind, is_js_array);
__ CallStub(&stub);
RecordSafepointWithRegisters(
instr->pointer_map(), 0, Safepoint::kLazyDeopt);
}
__ bind(&not_applicable);
}
void LCodeGen::DoTrapAllocationMemento(LTrapAllocationMemento* instr) {
Register object = ToRegister(instr->object());
Register temp = ToRegister(instr->temp());
Label no_memento_found;
__ TestJSArrayForAllocationMemento(object, temp, &no_memento_found);
DeoptimizeIf(eq, instr, Deoptimizer::kMementoFound);
__ bind(&no_memento_found);
}
void LCodeGen::DoStringAdd(LStringAdd* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
DCHECK(ToRegister(instr->left()).is(r1));
DCHECK(ToRegister(instr->right()).is(r0));
StringAddStub stub(isolate(),
instr->hydrogen()->flags(),
instr->hydrogen()->pretenure_flag());
CallCode(stub.GetCode(), RelocInfo::CODE_TARGET, instr);
}
void LCodeGen::DoStringCharCodeAt(LStringCharCodeAt* instr) {
class DeferredStringCharCodeAt FINAL : public LDeferredCode {
public:
DeferredStringCharCodeAt(LCodeGen* codegen, LStringCharCodeAt* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE { codegen()->DoDeferredStringCharCodeAt(instr_); }
LInstruction* instr() OVERRIDE { return instr_; }
private:
LStringCharCodeAt* instr_;
};
DeferredStringCharCodeAt* deferred =
new(zone()) DeferredStringCharCodeAt(this, instr);
StringCharLoadGenerator::Generate(masm(),
ToRegister(instr->string()),
ToRegister(instr->index()),
ToRegister(instr->result()),
deferred->entry());
__ bind(deferred->exit());
}
void LCodeGen::DoDeferredStringCharCodeAt(LStringCharCodeAt* instr) {
Register string = ToRegister(instr->string());
Register result = ToRegister(instr->result());
Register scratch = scratch0();
// TODO(3095996): Get rid of this. For now, we need to make the
// result register contain a valid pointer because it is already
// contained in the register pointer map.
__ mov(result, Operand::Zero());
PushSafepointRegistersScope scope(this);
__ push(string);
// Push the index as a smi. This is safe because of the checks in
// DoStringCharCodeAt above.
if (instr->index()->IsConstantOperand()) {
int const_index = ToInteger32(LConstantOperand::cast(instr->index()));
__ mov(scratch, Operand(Smi::FromInt(const_index)));
__ push(scratch);
} else {
Register index = ToRegister(instr->index());
__ SmiTag(index);
__ push(index);
}
CallRuntimeFromDeferred(Runtime::kStringCharCodeAtRT, 2, instr,
instr->context());
__ AssertSmi(r0);
__ SmiUntag(r0);
__ StoreToSafepointRegisterSlot(r0, result);
}
void LCodeGen::DoStringCharFromCode(LStringCharFromCode* instr) {
class DeferredStringCharFromCode FINAL : public LDeferredCode {
public:
DeferredStringCharFromCode(LCodeGen* codegen, LStringCharFromCode* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE {
codegen()->DoDeferredStringCharFromCode(instr_);
}
LInstruction* instr() OVERRIDE { return instr_; }
private:
LStringCharFromCode* instr_;
};
DeferredStringCharFromCode* deferred =
new(zone()) DeferredStringCharFromCode(this, instr);
DCHECK(instr->hydrogen()->value()->representation().IsInteger32());
Register char_code = ToRegister(instr->char_code());
Register result = ToRegister(instr->result());
DCHECK(!char_code.is(result));
__ cmp(char_code, Operand(String::kMaxOneByteCharCode));
__ b(hi, deferred->entry());
__ LoadRoot(result, Heap::kSingleCharacterStringCacheRootIndex);
__ add(result, result, Operand(char_code, LSL, kPointerSizeLog2));
__ ldr(result, FieldMemOperand(result, FixedArray::kHeaderSize));
__ LoadRoot(ip, Heap::kUndefinedValueRootIndex);
__ cmp(result, ip);
__ b(eq, deferred->entry());
__ bind(deferred->exit());
}
void LCodeGen::DoDeferredStringCharFromCode(LStringCharFromCode* instr) {
Register char_code = ToRegister(instr->char_code());
Register result = ToRegister(instr->result());
// TODO(3095996): Get rid of this. For now, we need to make the
// result register contain a valid pointer because it is already
// contained in the register pointer map.
__ mov(result, Operand::Zero());
PushSafepointRegistersScope scope(this);
__ SmiTag(char_code);
__ push(char_code);
CallRuntimeFromDeferred(Runtime::kCharFromCode, 1, instr, instr->context());
__ StoreToSafepointRegisterSlot(r0, result);
}
void LCodeGen::DoInteger32ToDouble(LInteger32ToDouble* instr) {
LOperand* input = instr->value();
DCHECK(input->IsRegister() || input->IsStackSlot());
LOperand* output = instr->result();
DCHECK(output->IsDoubleRegister());
SwVfpRegister single_scratch = double_scratch0().low();
if (input->IsStackSlot()) {
Register scratch = scratch0();
__ ldr(scratch, ToMemOperand(input));
__ vmov(single_scratch, scratch);
} else {
__ vmov(single_scratch, ToRegister(input));
}
__ vcvt_f64_s32(ToDoubleRegister(output), single_scratch);
}
void LCodeGen::DoUint32ToDouble(LUint32ToDouble* instr) {
LOperand* input = instr->value();
LOperand* output = instr->result();
SwVfpRegister flt_scratch = double_scratch0().low();
__ vmov(flt_scratch, ToRegister(input));
__ vcvt_f64_u32(ToDoubleRegister(output), flt_scratch);
}
void LCodeGen::DoNumberTagI(LNumberTagI* instr) {
class DeferredNumberTagI FINAL : public LDeferredCode {
public:
DeferredNumberTagI(LCodeGen* codegen, LNumberTagI* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE {
codegen()->DoDeferredNumberTagIU(instr_,
instr_->value(),
instr_->temp1(),
instr_->temp2(),
SIGNED_INT32);
}
LInstruction* instr() OVERRIDE { return instr_; }
private:
LNumberTagI* instr_;
};
Register src = ToRegister(instr->value());
Register dst = ToRegister(instr->result());
DeferredNumberTagI* deferred = new(zone()) DeferredNumberTagI(this, instr);
__ SmiTag(dst, src, SetCC);
__ b(vs, deferred->entry());
__ bind(deferred->exit());
}
void LCodeGen::DoNumberTagU(LNumberTagU* instr) {
class DeferredNumberTagU FINAL : public LDeferredCode {
public:
DeferredNumberTagU(LCodeGen* codegen, LNumberTagU* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE {
codegen()->DoDeferredNumberTagIU(instr_,
instr_->value(),
instr_->temp1(),
instr_->temp2(),
UNSIGNED_INT32);
}
LInstruction* instr() OVERRIDE { return instr_; }
private:
LNumberTagU* instr_;
};
Register input = ToRegister(instr->value());
Register result = ToRegister(instr->result());
DeferredNumberTagU* deferred = new(zone()) DeferredNumberTagU(this, instr);
__ cmp(input, Operand(Smi::kMaxValue));
__ b(hi, deferred->entry());
__ SmiTag(result, input);
__ bind(deferred->exit());
}
void LCodeGen::DoDeferredNumberTagIU(LInstruction* instr,
LOperand* value,
LOperand* temp1,
LOperand* temp2,
IntegerSignedness signedness) {
Label done, slow;
Register src = ToRegister(value);
Register dst = ToRegister(instr->result());
Register tmp1 = scratch0();
Register tmp2 = ToRegister(temp1);
Register tmp3 = ToRegister(temp2);
LowDwVfpRegister dbl_scratch = double_scratch0();
if (signedness == SIGNED_INT32) {
// There was overflow, so bits 30 and 31 of the original integer
// disagree. Try to allocate a heap number in new space and store
// the value in there. If that fails, call the runtime system.
if (dst.is(src)) {
__ SmiUntag(src, dst);
__ eor(src, src, Operand(0x80000000));
}
__ vmov(dbl_scratch.low(), src);
__ vcvt_f64_s32(dbl_scratch, dbl_scratch.low());
} else {
__ vmov(dbl_scratch.low(), src);
__ vcvt_f64_u32(dbl_scratch, dbl_scratch.low());
}
if (FLAG_inline_new) {
__ LoadRoot(tmp3, Heap::kHeapNumberMapRootIndex);
__ AllocateHeapNumber(dst, tmp1, tmp2, tmp3, &slow, DONT_TAG_RESULT);
__ b(&done);
}
// Slow case: Call the runtime system to do the number allocation.
__ bind(&slow);
{
// TODO(3095996): Put a valid pointer value in the stack slot where the
// result register is stored, as this register is in the pointer map, but
// contains an integer value.
__ mov(dst, Operand::Zero());
// Preserve the value of all registers.
PushSafepointRegistersScope scope(this);
// NumberTagI and NumberTagD use the context from the frame, rather than
// the environment's HContext or HInlinedContext value.
// They only call Runtime::kAllocateHeapNumber.
// The corresponding HChange instructions are added in a phase that does
// not have easy access to the local context.
__ ldr(cp, MemOperand(fp, StandardFrameConstants::kContextOffset));
__ CallRuntimeSaveDoubles(Runtime::kAllocateHeapNumber);
RecordSafepointWithRegisters(
instr->pointer_map(), 0, Safepoint::kNoLazyDeopt);
__ sub(r0, r0, Operand(kHeapObjectTag));
__ StoreToSafepointRegisterSlot(r0, dst);
}
// Done. Put the value in dbl_scratch into the value of the allocated heap
// number.
__ bind(&done);
__ vstr(dbl_scratch, dst, HeapNumber::kValueOffset);
__ add(dst, dst, Operand(kHeapObjectTag));
}
void LCodeGen::DoNumberTagD(LNumberTagD* instr) {
class DeferredNumberTagD FINAL : public LDeferredCode {
public:
DeferredNumberTagD(LCodeGen* codegen, LNumberTagD* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE { codegen()->DoDeferredNumberTagD(instr_); }
LInstruction* instr() OVERRIDE { return instr_; }
private:
LNumberTagD* instr_;
};
DwVfpRegister input_reg = ToDoubleRegister(instr->value());
Register scratch = scratch0();
Register reg = ToRegister(instr->result());
Register temp1 = ToRegister(instr->temp());
Register temp2 = ToRegister(instr->temp2());
DeferredNumberTagD* deferred = new(zone()) DeferredNumberTagD(this, instr);
if (FLAG_inline_new) {
__ LoadRoot(scratch, Heap::kHeapNumberMapRootIndex);
// We want the untagged address first for performance
__ AllocateHeapNumber(reg, temp1, temp2, scratch, deferred->entry(),
DONT_TAG_RESULT);
} else {
__ jmp(deferred->entry());
}
__ bind(deferred->exit());
__ vstr(input_reg, reg, HeapNumber::kValueOffset);
// Now that we have finished with the object's real address tag it
__ add(reg, reg, Operand(kHeapObjectTag));
}
void LCodeGen::DoDeferredNumberTagD(LNumberTagD* instr) {
// TODO(3095996): Get rid of this. For now, we need to make the
// result register contain a valid pointer because it is already
// contained in the register pointer map.
Register reg = ToRegister(instr->result());
__ mov(reg, Operand::Zero());
PushSafepointRegistersScope scope(this);
// NumberTagI and NumberTagD use the context from the frame, rather than
// the environment's HContext or HInlinedContext value.
// They only call Runtime::kAllocateHeapNumber.
// The corresponding HChange instructions are added in a phase that does
// not have easy access to the local context.
__ ldr(cp, MemOperand(fp, StandardFrameConstants::kContextOffset));
__ CallRuntimeSaveDoubles(Runtime::kAllocateHeapNumber);
RecordSafepointWithRegisters(
instr->pointer_map(), 0, Safepoint::kNoLazyDeopt);
__ sub(r0, r0, Operand(kHeapObjectTag));
__ StoreToSafepointRegisterSlot(r0, reg);
}
void LCodeGen::DoSmiTag(LSmiTag* instr) {
HChange* hchange = instr->hydrogen();
Register input = ToRegister(instr->value());
Register output = ToRegister(instr->result());
if (hchange->CheckFlag(HValue::kCanOverflow) &&
hchange->value()->CheckFlag(HValue::kUint32)) {
__ tst(input, Operand(0xc0000000));
DeoptimizeIf(ne, instr, Deoptimizer::kOverflow);
}
if (hchange->CheckFlag(HValue::kCanOverflow) &&
!hchange->value()->CheckFlag(HValue::kUint32)) {
__ SmiTag(output, input, SetCC);
DeoptimizeIf(vs, instr, Deoptimizer::kOverflow);
} else {
__ SmiTag(output, input);
}
}
void LCodeGen::DoSmiUntag(LSmiUntag* instr) {
Register input = ToRegister(instr->value());
Register result = ToRegister(instr->result());
if (instr->needs_check()) {
STATIC_ASSERT(kHeapObjectTag == 1);
// If the input is a HeapObject, SmiUntag will set the carry flag.
__ SmiUntag(result, input, SetCC);
DeoptimizeIf(cs, instr, Deoptimizer::kNotASmi);
} else {
__ SmiUntag(result, input);
}
}
void LCodeGen::EmitNumberUntagD(LNumberUntagD* instr, Register input_reg,
DwVfpRegister result_reg,
NumberUntagDMode mode) {
bool can_convert_undefined_to_nan =
instr->hydrogen()->can_convert_undefined_to_nan();
bool deoptimize_on_minus_zero = instr->hydrogen()->deoptimize_on_minus_zero();
Register scratch = scratch0();
SwVfpRegister flt_scratch = double_scratch0().low();
DCHECK(!result_reg.is(double_scratch0()));
Label convert, load_smi, done;
if (mode == NUMBER_CANDIDATE_IS_ANY_TAGGED) {
// Smi check.
__ UntagAndJumpIfSmi(scratch, input_reg, &load_smi);
// Heap number map check.
__ ldr(scratch, FieldMemOperand(input_reg, HeapObject::kMapOffset));
__ LoadRoot(ip, Heap::kHeapNumberMapRootIndex);
__ cmp(scratch, Operand(ip));
if (can_convert_undefined_to_nan) {
__ b(ne, &convert);
} else {
DeoptimizeIf(ne, instr, Deoptimizer::kNotAHeapNumber);
}
// load heap number
__ vldr(result_reg, input_reg, HeapNumber::kValueOffset - kHeapObjectTag);
if (deoptimize_on_minus_zero) {
__ VmovLow(scratch, result_reg);
__ cmp(scratch, Operand::Zero());
__ b(ne, &done);
__ VmovHigh(scratch, result_reg);
__ cmp(scratch, Operand(HeapNumber::kSignMask));
DeoptimizeIf(eq, instr, Deoptimizer::kMinusZero);
}
__ jmp(&done);
if (can_convert_undefined_to_nan) {
__ bind(&convert);
// Convert undefined (and hole) to NaN.
__ LoadRoot(ip, Heap::kUndefinedValueRootIndex);
__ cmp(input_reg, Operand(ip));
DeoptimizeIf(ne, instr, Deoptimizer::kNotAHeapNumberUndefined);
__ LoadRoot(scratch, Heap::kNanValueRootIndex);
__ vldr(result_reg, scratch, HeapNumber::kValueOffset - kHeapObjectTag);
__ jmp(&done);
}
} else {
__ SmiUntag(scratch, input_reg);
DCHECK(mode == NUMBER_CANDIDATE_IS_SMI);
}
// Smi to double register conversion
__ bind(&load_smi);
// scratch: untagged value of input_reg
__ vmov(flt_scratch, scratch);
__ vcvt_f64_s32(result_reg, flt_scratch);
__ bind(&done);
}
void LCodeGen::DoDeferredTaggedToI(LTaggedToI* instr) {
Register input_reg = ToRegister(instr->value());
Register scratch1 = scratch0();
Register scratch2 = ToRegister(instr->temp());
LowDwVfpRegister double_scratch = double_scratch0();
DwVfpRegister double_scratch2 = ToDoubleRegister(instr->temp2());
DCHECK(!scratch1.is(input_reg) && !scratch1.is(scratch2));
DCHECK(!scratch2.is(input_reg) && !scratch2.is(scratch1));
Label done;
// The input was optimistically untagged; revert it.
// The carry flag is set when we reach this deferred code as we just executed
// SmiUntag(heap_object, SetCC)
STATIC_ASSERT(kHeapObjectTag == 1);
__ adc(scratch2, input_reg, Operand(input_reg));
// Heap number map check.
__ ldr(scratch1, FieldMemOperand(scratch2, HeapObject::kMapOffset));
__ LoadRoot(ip, Heap::kHeapNumberMapRootIndex);
__ cmp(scratch1, Operand(ip));
if (instr->truncating()) {
// Performs a truncating conversion of a floating point number as used by
// the JS bitwise operations.
Label no_heap_number, check_bools, check_false;
__ b(ne, &no_heap_number);
__ TruncateHeapNumberToI(input_reg, scratch2);
__ b(&done);
// Check for Oddballs. Undefined/False is converted to zero and True to one
// for truncating conversions.
__ bind(&no_heap_number);
__ LoadRoot(ip, Heap::kUndefinedValueRootIndex);
__ cmp(scratch2, Operand(ip));
__ b(ne, &check_bools);
__ mov(input_reg, Operand::Zero());
__ b(&done);
__ bind(&check_bools);
__ LoadRoot(ip, Heap::kTrueValueRootIndex);
__ cmp(scratch2, Operand(ip));
__ b(ne, &check_false);
__ mov(input_reg, Operand(1));
__ b(&done);
__ bind(&check_false);
__ LoadRoot(ip, Heap::kFalseValueRootIndex);
__ cmp(scratch2, Operand(ip));
DeoptimizeIf(ne, instr, Deoptimizer::kNotAHeapNumberUndefinedBoolean);
__ mov(input_reg, Operand::Zero());
} else {
DeoptimizeIf(ne, instr, Deoptimizer::kNotAHeapNumber);
__ sub(ip, scratch2, Operand(kHeapObjectTag));
__ vldr(double_scratch2, ip, HeapNumber::kValueOffset);
__ TryDoubleToInt32Exact(input_reg, double_scratch2, double_scratch);
DeoptimizeIf(ne, instr, Deoptimizer::kLostPrecisionOrNaN);
if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
__ cmp(input_reg, Operand::Zero());
__ b(ne, &done);
__ VmovHigh(scratch1, double_scratch2);
__ tst(scratch1, Operand(HeapNumber::kSignMask));
DeoptimizeIf(ne, instr, Deoptimizer::kMinusZero);
}
}
__ bind(&done);
}
void LCodeGen::DoTaggedToI(LTaggedToI* instr) {
class DeferredTaggedToI FINAL : public LDeferredCode {
public:
DeferredTaggedToI(LCodeGen* codegen, LTaggedToI* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE { codegen()->DoDeferredTaggedToI(instr_); }
LInstruction* instr() OVERRIDE { return instr_; }
private:
LTaggedToI* instr_;
};
LOperand* input = instr->value();
DCHECK(input->IsRegister());
DCHECK(input->Equals(instr->result()));
Register input_reg = ToRegister(input);
if (instr->hydrogen()->value()->representation().IsSmi()) {
__ SmiUntag(input_reg);
} else {
DeferredTaggedToI* deferred = new(zone()) DeferredTaggedToI(this, instr);
// Optimistically untag the input.
// If the input is a HeapObject, SmiUntag will set the carry flag.
__ SmiUntag(input_reg, SetCC);
// Branch to deferred code if the input was tagged.
// The deferred code will take care of restoring the tag.
__ b(cs, deferred->entry());
__ bind(deferred->exit());
}
}
void LCodeGen::DoNumberUntagD(LNumberUntagD* instr) {
LOperand* input = instr->value();
DCHECK(input->IsRegister());
LOperand* result = instr->result();
DCHECK(result->IsDoubleRegister());
Register input_reg = ToRegister(input);
DwVfpRegister result_reg = ToDoubleRegister(result);
HValue* value = instr->hydrogen()->value();
NumberUntagDMode mode = value->representation().IsSmi()
? NUMBER_CANDIDATE_IS_SMI : NUMBER_CANDIDATE_IS_ANY_TAGGED;
EmitNumberUntagD(instr, input_reg, result_reg, mode);
}
void LCodeGen::DoDoubleToI(LDoubleToI* instr) {
Register result_reg = ToRegister(instr->result());
Register scratch1 = scratch0();
DwVfpRegister double_input = ToDoubleRegister(instr->value());
LowDwVfpRegister double_scratch = double_scratch0();
if (instr->truncating()) {
__ TruncateDoubleToI(result_reg, double_input);
} else {
__ TryDoubleToInt32Exact(result_reg, double_input, double_scratch);
// Deoptimize if the input wasn't a int32 (inside a double).
DeoptimizeIf(ne, instr, Deoptimizer::kLostPrecisionOrNaN);
if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
Label done;
__ cmp(result_reg, Operand::Zero());
__ b(ne, &done);
__ VmovHigh(scratch1, double_input);
__ tst(scratch1, Operand(HeapNumber::kSignMask));
DeoptimizeIf(ne, instr, Deoptimizer::kMinusZero);
__ bind(&done);
}
}
}
void LCodeGen::DoDoubleToSmi(LDoubleToSmi* instr) {
Register result_reg = ToRegister(instr->result());
Register scratch1 = scratch0();
DwVfpRegister double_input = ToDoubleRegister(instr->value());
LowDwVfpRegister double_scratch = double_scratch0();
if (instr->truncating()) {
__ TruncateDoubleToI(result_reg, double_input);
} else {
__ TryDoubleToInt32Exact(result_reg, double_input, double_scratch);
// Deoptimize if the input wasn't a int32 (inside a double).
DeoptimizeIf(ne, instr, Deoptimizer::kLostPrecisionOrNaN);
if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
Label done;
__ cmp(result_reg, Operand::Zero());
__ b(ne, &done);
__ VmovHigh(scratch1, double_input);
__ tst(scratch1, Operand(HeapNumber::kSignMask));
DeoptimizeIf(ne, instr, Deoptimizer::kMinusZero);
__ bind(&done);
}
}
__ SmiTag(result_reg, SetCC);
DeoptimizeIf(vs, instr, Deoptimizer::kOverflow);
}
void LCodeGen::DoCheckSmi(LCheckSmi* instr) {
LOperand* input = instr->value();
__ SmiTst(ToRegister(input));
DeoptimizeIf(ne, instr, Deoptimizer::kNotASmi);
}
void LCodeGen::DoCheckNonSmi(LCheckNonSmi* instr) {
if (!instr->hydrogen()->value()->type().IsHeapObject()) {
LOperand* input = instr->value();
__ SmiTst(ToRegister(input));
DeoptimizeIf(eq, instr, Deoptimizer::kSmi);
}
}
void LCodeGen::DoCheckInstanceType(LCheckInstanceType* instr) {
Register input = ToRegister(instr->value());
Register scratch = scratch0();
__ ldr(scratch, FieldMemOperand(input, HeapObject::kMapOffset));
__ ldrb(scratch, FieldMemOperand(scratch, Map::kInstanceTypeOffset));
if (instr->hydrogen()->is_interval_check()) {
InstanceType first;
InstanceType last;
instr->hydrogen()->GetCheckInterval(&first, &last);
__ cmp(scratch, Operand(first));
// If there is only one type in the interval check for equality.
if (first == last) {
DeoptimizeIf(ne, instr, Deoptimizer::kWrongInstanceType);
} else {
DeoptimizeIf(lo, instr, Deoptimizer::kWrongInstanceType);
// Omit check for the last type.
if (last != LAST_TYPE) {
__ cmp(scratch, Operand(last));
DeoptimizeIf(hi, instr, Deoptimizer::kWrongInstanceType);
}
}
} else {
uint8_t mask;
uint8_t tag;
instr->hydrogen()->GetCheckMaskAndTag(&mask, &tag);
if (base::bits::IsPowerOfTwo32(mask)) {
DCHECK(tag == 0 || base::bits::IsPowerOfTwo32(tag));
__ tst(scratch, Operand(mask));
DeoptimizeIf(tag == 0 ? ne : eq, instr, Deoptimizer::kWrongInstanceType);
} else {
__ and_(scratch, scratch, Operand(mask));
__ cmp(scratch, Operand(tag));
DeoptimizeIf(ne, instr, Deoptimizer::kWrongInstanceType);
}
}
}
void LCodeGen::DoCheckValue(LCheckValue* instr) {
Register reg = ToRegister(instr->value());
Handle<HeapObject> object = instr->hydrogen()->object().handle();
AllowDeferredHandleDereference smi_check;
if (isolate()->heap()->InNewSpace(*object)) {
Register reg = ToRegister(instr->value());
Handle<Cell> cell = isolate()->factory()->NewCell(object);
__ mov(ip, Operand(cell));
__ ldr(ip, FieldMemOperand(ip, Cell::kValueOffset));
__ cmp(reg, ip);
} else {
__ cmp(reg, Operand(object));
}
DeoptimizeIf(ne, instr, Deoptimizer::kValueMismatch);
}
void LCodeGen::DoDeferredInstanceMigration(LCheckMaps* instr, Register object) {
{
PushSafepointRegistersScope scope(this);
__ push(object);
__ mov(cp, Operand::Zero());
__ CallRuntimeSaveDoubles(Runtime::kTryMigrateInstance);
RecordSafepointWithRegisters(
instr->pointer_map(), 1, Safepoint::kNoLazyDeopt);
__ StoreToSafepointRegisterSlot(r0, scratch0());
}
__ tst(scratch0(), Operand(kSmiTagMask));
DeoptimizeIf(eq, instr, Deoptimizer::kInstanceMigrationFailed);
}
void LCodeGen::DoCheckMaps(LCheckMaps* instr) {
class DeferredCheckMaps FINAL : public LDeferredCode {
public:
DeferredCheckMaps(LCodeGen* codegen, LCheckMaps* instr, Register object)
: LDeferredCode(codegen), instr_(instr), object_(object) {
SetExit(check_maps());
}
void Generate() OVERRIDE {
codegen()->DoDeferredInstanceMigration(instr_, object_);
}
Label* check_maps() { return &check_maps_; }
LInstruction* instr() OVERRIDE { return instr_; }
private:
LCheckMaps* instr_;
Label check_maps_;
Register object_;
};
if (instr->hydrogen()->IsStabilityCheck()) {
const UniqueSet<Map>* maps = instr->hydrogen()->maps();
for (int i = 0; i < maps->size(); ++i) {
AddStabilityDependency(maps->at(i).handle());
}
return;
}
Register map_reg = scratch0();
LOperand* input = instr->value();
DCHECK(input->IsRegister());
Register reg = ToRegister(input);
__ ldr(map_reg, FieldMemOperand(reg, HeapObject::kMapOffset));
DeferredCheckMaps* deferred = NULL;
if (instr->hydrogen()->HasMigrationTarget()) {
deferred = new(zone()) DeferredCheckMaps(this, instr, reg);
__ bind(deferred->check_maps());
}
const UniqueSet<Map>* maps = instr->hydrogen()->maps();
Label success;
for (int i = 0; i < maps->size() - 1; i++) {
Handle<Map> map = maps->at(i).handle();
__ CompareMap(map_reg, map, &success);
__ b(eq, &success);
}
Handle<Map> map = maps->at(maps->size() - 1).handle();
__ CompareMap(map_reg, map, &success);
if (instr->hydrogen()->HasMigrationTarget()) {
__ b(ne, deferred->entry());
} else {
DeoptimizeIf(ne, instr, Deoptimizer::kWrongMap);
}
__ bind(&success);
}
void LCodeGen::DoClampDToUint8(LClampDToUint8* instr) {
DwVfpRegister value_reg = ToDoubleRegister(instr->unclamped());
Register result_reg = ToRegister(instr->result());
__ ClampDoubleToUint8(result_reg, value_reg, double_scratch0());
}
void LCodeGen::DoClampIToUint8(LClampIToUint8* instr) {
Register unclamped_reg = ToRegister(instr->unclamped());
Register result_reg = ToRegister(instr->result());
__ ClampUint8(result_reg, unclamped_reg);
}
void LCodeGen::DoClampTToUint8(LClampTToUint8* instr) {
Register scratch = scratch0();
Register input_reg = ToRegister(instr->unclamped());
Register result_reg = ToRegister(instr->result());
DwVfpRegister temp_reg = ToDoubleRegister(instr->temp());
Label is_smi, done, heap_number;
// Both smi and heap number cases are handled.
__ UntagAndJumpIfSmi(result_reg, input_reg, &is_smi);
// Check for heap number
__ ldr(scratch, FieldMemOperand(input_reg, HeapObject::kMapOffset));
__ cmp(scratch, Operand(factory()->heap_number_map()));
__ b(eq, &heap_number);
// Check for undefined. Undefined is converted to zero for clamping
// conversions.
__ cmp(input_reg, Operand(factory()->undefined_value()));
DeoptimizeIf(ne, instr, Deoptimizer::kNotAHeapNumberUndefined);
__ mov(result_reg, Operand::Zero());
__ jmp(&done);
// Heap number
__ bind(&heap_number);
__ vldr(temp_reg, FieldMemOperand(input_reg, HeapNumber::kValueOffset));
__ ClampDoubleToUint8(result_reg, temp_reg, double_scratch0());
__ jmp(&done);
// smi
__ bind(&is_smi);
__ ClampUint8(result_reg, result_reg);
__ bind(&done);
}
void LCodeGen::DoDoubleBits(LDoubleBits* instr) {
DwVfpRegister value_reg = ToDoubleRegister(instr->value());
Register result_reg = ToRegister(instr->result());
if (instr->hydrogen()->bits() == HDoubleBits::HIGH) {
__ VmovHigh(result_reg, value_reg);
} else {
__ VmovLow(result_reg, value_reg);
}
}
void LCodeGen::DoConstructDouble(LConstructDouble* instr) {
Register hi_reg = ToRegister(instr->hi());
Register lo_reg = ToRegister(instr->lo());
DwVfpRegister result_reg = ToDoubleRegister(instr->result());
__ VmovHigh(result_reg, hi_reg);
__ VmovLow(result_reg, lo_reg);
}
void LCodeGen::DoAllocate(LAllocate* instr) {
class DeferredAllocate FINAL : public LDeferredCode {
public:
DeferredAllocate(LCodeGen* codegen, LAllocate* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE { codegen()->DoDeferredAllocate(instr_); }
LInstruction* instr() OVERRIDE { return instr_; }
private:
LAllocate* instr_;
};
DeferredAllocate* deferred =
new(zone()) DeferredAllocate(this, instr);
Register result = ToRegister(instr->result());
Register scratch = ToRegister(instr->temp1());
Register scratch2 = ToRegister(instr->temp2());
// Allocate memory for the object.
AllocationFlags flags = TAG_OBJECT;
if (instr->hydrogen()->MustAllocateDoubleAligned()) {
flags = static_cast<AllocationFlags>(flags | DOUBLE_ALIGNMENT);
}
if (instr->hydrogen()->IsOldPointerSpaceAllocation()) {
DCHECK(!instr->hydrogen()->IsOldDataSpaceAllocation());
DCHECK(!instr->hydrogen()->IsNewSpaceAllocation());
flags = static_cast<AllocationFlags>(flags | PRETENURE_OLD_POINTER_SPACE);
} else if (instr->hydrogen()->IsOldDataSpaceAllocation()) {
DCHECK(!instr->hydrogen()->IsNewSpaceAllocation());
flags = static_cast<AllocationFlags>(flags | PRETENURE_OLD_DATA_SPACE);
}
if (instr->size()->IsConstantOperand()) {
int32_t size = ToInteger32(LConstantOperand::cast(instr->size()));
if (size <= Page::kMaxRegularHeapObjectSize) {
__ Allocate(size, result, scratch, scratch2, deferred->entry(), flags);
} else {
__ jmp(deferred->entry());
}
} else {
Register size = ToRegister(instr->size());
__ Allocate(size, result, scratch, scratch2, deferred->entry(), flags);
}
__ bind(deferred->exit());
if (instr->hydrogen()->MustPrefillWithFiller()) {
STATIC_ASSERT(kHeapObjectTag == 1);
if (instr->size()->IsConstantOperand()) {
int32_t size = ToInteger32(LConstantOperand::cast(instr->size()));
__ mov(scratch, Operand(size - kHeapObjectTag));
} else {
__ sub(scratch, ToRegister(instr->size()), Operand(kHeapObjectTag));
}
__ mov(scratch2, Operand(isolate()->factory()->one_pointer_filler_map()));
Label loop;
__ bind(&loop);
__ sub(scratch, scratch, Operand(kPointerSize), SetCC);
__ str(scratch2, MemOperand(result, scratch));
__ b(ge, &loop);
}
}
void LCodeGen::DoDeferredAllocate(LAllocate* instr) {
Register result = ToRegister(instr->result());
// TODO(3095996): Get rid of this. For now, we need to make the
// result register contain a valid pointer because it is already
// contained in the register pointer map.
__ mov(result, Operand(Smi::FromInt(0)));
PushSafepointRegistersScope scope(this);
if (instr->size()->IsRegister()) {
Register size = ToRegister(instr->size());
DCHECK(!size.is(result));
__ SmiTag(size);
__ push(size);
} else {
int32_t size = ToInteger32(LConstantOperand::cast(instr->size()));
if (size >= 0 && size <= Smi::kMaxValue) {
__ Push(Smi::FromInt(size));
} else {
// We should never get here at runtime => abort
__ stop("invalid allocation size");
return;
}
}
int flags = AllocateDoubleAlignFlag::encode(
instr->hydrogen()->MustAllocateDoubleAligned());
if (instr->hydrogen()->IsOldPointerSpaceAllocation()) {
DCHECK(!instr->hydrogen()->IsOldDataSpaceAllocation());
DCHECK(!instr->hydrogen()->IsNewSpaceAllocation());
flags = AllocateTargetSpace::update(flags, OLD_POINTER_SPACE);
} else if (instr->hydrogen()->IsOldDataSpaceAllocation()) {
DCHECK(!instr->hydrogen()->IsNewSpaceAllocation());
flags = AllocateTargetSpace::update(flags, OLD_DATA_SPACE);
} else {
flags = AllocateTargetSpace::update(flags, NEW_SPACE);
}
__ Push(Smi::FromInt(flags));
CallRuntimeFromDeferred(
Runtime::kAllocateInTargetSpace, 2, instr, instr->context());
__ StoreToSafepointRegisterSlot(r0, result);
}
void LCodeGen::DoToFastProperties(LToFastProperties* instr) {
DCHECK(ToRegister(instr->value()).is(r0));
__ push(r0);
CallRuntime(Runtime::kToFastProperties, 1, instr);
}
void LCodeGen::DoRegExpLiteral(LRegExpLiteral* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
Label materialized;
// Registers will be used as follows:
// r6 = literals array.
// r1 = regexp literal.
// r0 = regexp literal clone.
// r2-5 are used as temporaries.
int literal_offset =
FixedArray::OffsetOfElementAt(instr->hydrogen()->literal_index());
__ Move(r6, instr->hydrogen()->literals());
__ ldr(r1, FieldMemOperand(r6, literal_offset));
__ LoadRoot(ip, Heap::kUndefinedValueRootIndex);
__ cmp(r1, ip);
__ b(ne, &materialized);
// Create regexp literal using runtime function
// Result will be in r0.
__ mov(r5, Operand(Smi::FromInt(instr->hydrogen()->literal_index())));
__ mov(r4, Operand(instr->hydrogen()->pattern()));
__ mov(r3, Operand(instr->hydrogen()->flags()));
__ Push(r6, r5, r4, r3);
CallRuntime(Runtime::kMaterializeRegExpLiteral, 4, instr);
__ mov(r1, r0);
__ bind(&materialized);
int size = JSRegExp::kSize + JSRegExp::kInObjectFieldCount * kPointerSize;
Label allocated, runtime_allocate;
__ Allocate(size, r0, r2, r3, &runtime_allocate, TAG_OBJECT);
__ jmp(&allocated);
__ bind(&runtime_allocate);
__ mov(r0, Operand(Smi::FromInt(size)));
__ Push(r1, r0);
CallRuntime(Runtime::kAllocateInNewSpace, 1, instr);
__ pop(r1);
__ bind(&allocated);
// Copy the content into the newly allocated memory.
__ CopyFields(r0, r1, double_scratch0(), size / kPointerSize);
}
void LCodeGen::DoFunctionLiteral(LFunctionLiteral* instr) {
DCHECK(ToRegister(instr->context()).is(cp));
// Use the fast case closure allocation code that allocates in new
// space for nested functions that don't need literals cloning.
bool pretenure = instr->hydrogen()->pretenure();
if (!pretenure && instr->hydrogen()->has_no_literals()) {
FastNewClosureStub stub(isolate(), instr->hydrogen()->language_mode(),
instr->hydrogen()->kind());
__ mov(r2, Operand(instr->hydrogen()->shared_info()));
CallCode(stub.GetCode(), RelocInfo::CODE_TARGET, instr);
} else {
__ mov(r2, Operand(instr->hydrogen()->shared_info()));
__ mov(r1, Operand(pretenure ? factory()->true_value()
: factory()->false_value()));
__ Push(cp, r2, r1);
CallRuntime(Runtime::kNewClosure, 3, instr);
}
}
void LCodeGen::DoTypeof(LTypeof* instr) {
Register input = ToRegister(instr->value());
__ push(input);
CallRuntime(Runtime::kTypeof, 1, instr);
}
void LCodeGen::DoTypeofIsAndBranch(LTypeofIsAndBranch* instr) {
Register input = ToRegister(instr->value());
Condition final_branch_condition = EmitTypeofIs(instr->TrueLabel(chunk_),
instr->FalseLabel(chunk_),
input,
instr->type_literal());
if (final_branch_condition != kNoCondition) {
EmitBranch(instr, final_branch_condition);
}
}
Condition LCodeGen::EmitTypeofIs(Label* true_label,
Label* false_label,
Register input,
Handle<String> type_name) {
Condition final_branch_condition = kNoCondition;
Register scratch = scratch0();
Factory* factory = isolate()->factory();
if (String::Equals(type_name, factory->number_string())) {
__ JumpIfSmi(input, true_label);
__ ldr(scratch, FieldMemOperand(input, HeapObject::kMapOffset));
__ CompareRoot(scratch, Heap::kHeapNumberMapRootIndex);
final_branch_condition = eq;
} else if (String::Equals(type_name, factory->string_string())) {
__ JumpIfSmi(input, false_label);
__ CompareObjectType(input, scratch, no_reg, FIRST_NONSTRING_TYPE);
__ b(ge, false_label);
__ ldrb(scratch, FieldMemOperand(scratch, Map::kBitFieldOffset));
__ tst(scratch, Operand(1 << Map::kIsUndetectable));
final_branch_condition = eq;
} else if (String::Equals(type_name, factory->symbol_string())) {
__ JumpIfSmi(input, false_label);
__ CompareObjectType(input, scratch, no_reg, SYMBOL_TYPE);
final_branch_condition = eq;
} else if (String::Equals(type_name, factory->boolean_string())) {
__ CompareRoot(input, Heap::kTrueValueRootIndex);
__ b(eq, true_label);
__ CompareRoot(input, Heap::kFalseValueRootIndex);
final_branch_condition = eq;
} else if (String::Equals(type_name, factory->undefined_string())) {
__ CompareRoot(input, Heap::kUndefinedValueRootIndex);
__ b(eq, true_label);
__ JumpIfSmi(input, false_label);
// Check for undetectable objects => true.
__ ldr(scratch, FieldMemOperand(input, HeapObject::kMapOffset));
__ ldrb(scratch, FieldMemOperand(scratch, Map::kBitFieldOffset));
__ tst(scratch, Operand(1 << Map::kIsUndetectable));
final_branch_condition = ne;
} else if (String::Equals(type_name, factory->function_string())) {
STATIC_ASSERT(NUM_OF_CALLABLE_SPEC_OBJECT_TYPES == 2);
Register type_reg = scratch;
__ JumpIfSmi(input, false_label);
__ CompareObjectType(input, scratch, type_reg, JS_FUNCTION_TYPE);
__ b(eq, true_label);
__ cmp(type_reg, Operand(JS_FUNCTION_PROXY_TYPE));
final_branch_condition = eq;
} else if (String::Equals(type_name, factory->object_string())) {
Register map = scratch;
__ JumpIfSmi(input, false_label);
__ CompareRoot(input, Heap::kNullValueRootIndex);
__ b(eq, true_label);
__ CheckObjectTypeRange(input,
map,
FIRST_NONCALLABLE_SPEC_OBJECT_TYPE,
LAST_NONCALLABLE_SPEC_OBJECT_TYPE,
false_label);
// Check for undetectable objects => false.
__ ldrb(scratch, FieldMemOperand(map, Map::kBitFieldOffset));
__ tst(scratch, Operand(1 << Map::kIsUndetectable));
final_branch_condition = eq;
} else {
__ b(false_label);
}
return final_branch_condition;
}
void LCodeGen::DoIsConstructCallAndBranch(LIsConstructCallAndBranch* instr) {
Register temp1 = ToRegister(instr->temp());
EmitIsConstructCall(temp1, scratch0());
EmitBranch(instr, eq);
}
void LCodeGen::EmitIsConstructCall(Register temp1, Register temp2) {
DCHECK(!temp1.is(temp2));
// Get the frame pointer for the calling frame.
__ ldr(temp1, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
// Skip the arguments adaptor frame if it exists.
__ ldr(temp2, MemOperand(temp1, StandardFrameConstants::kContextOffset));
__ cmp(temp2, Operand(Smi::FromInt(StackFrame::ARGUMENTS_ADAPTOR)));
__ ldr(temp1, MemOperand(temp1, StandardFrameConstants::kCallerFPOffset), eq);
// Check the marker in the calling frame.
__ ldr(temp1, MemOperand(temp1, StandardFrameConstants::kMarkerOffset));
__ cmp(temp1, Operand(Smi::FromInt(StackFrame::CONSTRUCT)));
}
void LCodeGen::EnsureSpaceForLazyDeopt(int space_needed) {
if (!info()->IsStub()) {
// Ensure that we have enough space after the previous lazy-bailout
// instruction for patching the code here.
int current_pc = masm()->pc_offset();
if (current_pc < last_lazy_deopt_pc_ + space_needed) {
// Block literal pool emission for duration of padding.
Assembler::BlockConstPoolScope block_const_pool(masm());
int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
DCHECK_EQ(0, padding_size % Assembler::kInstrSize);
while (padding_size > 0) {
__ nop();
padding_size -= Assembler::kInstrSize;
}
}
}
last_lazy_deopt_pc_ = masm()->pc_offset();
}
void LCodeGen::DoLazyBailout(LLazyBailout* instr) {
last_lazy_deopt_pc_ = masm()->pc_offset();
DCHECK(instr->HasEnvironment());
LEnvironment* env = instr->environment();
RegisterEnvironmentForDeoptimization(env, Safepoint::kLazyDeopt);
safepoints_.RecordLazyDeoptimizationIndex(env->deoptimization_index());
}
void LCodeGen::DoDeoptimize(LDeoptimize* instr) {
Deoptimizer::BailoutType type = instr->hydrogen()->type();
// TODO(danno): Stubs expect all deopts to be lazy for historical reasons (the
// needed return address), even though the implementation of LAZY and EAGER is
// now identical. When LAZY is eventually completely folded into EAGER, remove
// the special case below.
if (info()->IsStub() && type == Deoptimizer::EAGER) {
type = Deoptimizer::LAZY;
}
DeoptimizeIf(al, instr, instr->hydrogen()->reason(), type);
}
void LCodeGen::DoDummy(LDummy* instr) {
// Nothing to see here, move on!
}
void LCodeGen::DoDummyUse(LDummyUse* instr) {
// Nothing to see here, move on!
}
void LCodeGen::DoDeferredStackCheck(LStackCheck* instr) {
PushSafepointRegistersScope scope(this);
LoadContextFromDeferred(instr->context());
__ CallRuntimeSaveDoubles(Runtime::kStackGuard);
RecordSafepointWithLazyDeopt(
instr, RECORD_SAFEPOINT_WITH_REGISTERS_AND_NO_ARGUMENTS);
DCHECK(instr->HasEnvironment());
LEnvironment* env = instr->environment();
safepoints_.RecordLazyDeoptimizationIndex(env->deoptimization_index());
}
void LCodeGen::DoStackCheck(LStackCheck* instr) {
class DeferredStackCheck FINAL : public LDeferredCode {
public:
DeferredStackCheck(LCodeGen* codegen, LStackCheck* instr)
: LDeferredCode(codegen), instr_(instr) { }
void Generate() OVERRIDE { codegen()->DoDeferredStackCheck(instr_); }
LInstruction* instr() OVERRIDE { return instr_; }
private:
LStackCheck* instr_;
};
DCHECK(instr->HasEnvironment());
LEnvironment* env = instr->environment();
// There is no LLazyBailout instruction for stack-checks. We have to
// prepare for lazy deoptimization explicitly here.
if (instr->hydrogen()->is_function_entry()) {
// Perform stack overflow check.
Label done;
__ LoadRoot(ip, Heap::kStackLimitRootIndex);
__ cmp(sp, Operand(ip));
__ b(hs, &done);
Handle<Code> stack_check = isolate()->builtins()->StackCheck();
PredictableCodeSizeScope predictable(masm(),
CallCodeSize(stack_check, RelocInfo::CODE_TARGET));
DCHECK(instr->context()->IsRegister());
DCHECK(ToRegister(instr->context()).is(cp));
CallCode(stack_check, RelocInfo::CODE_TARGET, instr);
__ bind(&done);
} else {
DCHECK(instr->hydrogen()->is_backwards_branch());
// Perform stack overflow check if this goto needs it before jumping.
DeferredStackCheck* deferred_stack_check =
new(zone()) DeferredStackCheck(this, instr);
__ LoadRoot(ip, Heap::kStackLimitRootIndex);
__ cmp(sp, Operand(ip));
__ b(lo, deferred_stack_check->entry());
EnsureSpaceForLazyDeopt(Deoptimizer::patch_size());
__ bind(instr->done_label());
deferred_stack_check->SetExit(instr->done_label());
RegisterEnvironmentForDeoptimization(env, Safepoint::kLazyDeopt);
// Don't record a deoptimization index for the safepoint here.
// This will be done explicitly when emitting call and the safepoint in
// the deferred code.
}
}
void LCodeGen::DoOsrEntry(LOsrEntry* instr) {
// This is a pseudo-instruction that ensures that the environment here is
// properly registered for deoptimization and records the assembler's PC
// offset.
LEnvironment* environment = instr->environment();
// If the environment were already registered, we would have no way of
// backpatching it with the spill slot operands.
DCHECK(!environment->HasBeenRegistered());
RegisterEnvironmentForDeoptimization(environment, Safepoint::kNoLazyDeopt);
GenerateOsrPrologue();
}
void LCodeGen::DoForInPrepareMap(LForInPrepareMap* instr) {
__ LoadRoot(ip, Heap::kUndefinedValueRootIndex);
__ cmp(r0, ip);
DeoptimizeIf(eq, instr, Deoptimizer::kUndefined);
Register null_value = r5;
__ LoadRoot(null_value, Heap::kNullValueRootIndex);
__ cmp(r0, null_value);
DeoptimizeIf(eq, instr, Deoptimizer::kNull);
__ SmiTst(r0);
DeoptimizeIf(eq, instr, Deoptimizer::kSmi);
STATIC_ASSERT(FIRST_JS_PROXY_TYPE == FIRST_SPEC_OBJECT_TYPE);
__ CompareObjectType(r0, r1, r1, LAST_JS_PROXY_TYPE);
DeoptimizeIf(le, instr, Deoptimizer::kWrongInstanceType);
Label use_cache, call_runtime;
__ CheckEnumCache(null_value, &call_runtime);
__ ldr(r0, FieldMemOperand(r0, HeapObject::kMapOffset));
__ b(&use_cache);
// Get the set of properties to enumerate.
__ bind(&call_runtime);
__ push(r0);
CallRuntime(Runtime::kGetPropertyNamesFast, 1, instr);
__ ldr(r1, FieldMemOperand(r0, HeapObject::kMapOffset));
__ LoadRoot(ip, Heap::kMetaMapRootIndex);
__ cmp(r1, ip);
DeoptimizeIf(ne, instr, Deoptimizer::kWrongMap);
__ bind(&use_cache);
}
void LCodeGen::DoForInCacheArray(LForInCacheArray* instr) {
Register map = ToRegister(instr->map());
Register result = ToRegister(instr->result());
Label load_cache, done;
__ EnumLength(result, map);
__ cmp(result, Operand(Smi::FromInt(0)));
__ b(ne, &load_cache);
__ mov(result, Operand(isolate()->factory()->empty_fixed_array()));
__ jmp(&done);
__ bind(&load_cache);
__ LoadInstanceDescriptors(map, result);
__ ldr(result,
FieldMemOperand(result, DescriptorArray::kEnumCacheOffset));
__ ldr(result,
FieldMemOperand(result, FixedArray::SizeFor(instr->idx())));
__ cmp(result, Operand::Zero());
DeoptimizeIf(eq, instr, Deoptimizer::kNoCache);
__ bind(&done);
}
void LCodeGen::DoCheckMapValue(LCheckMapValue* instr) {
Register object = ToRegister(instr->value());
Register map = ToRegister(instr->map());
__ ldr(scratch0(), FieldMemOperand(object, HeapObject::kMapOffset));
__ cmp(map, scratch0());
DeoptimizeIf(ne, instr, Deoptimizer::kWrongMap);
}
void LCodeGen::DoDeferredLoadMutableDouble(LLoadFieldByIndex* instr,
Register result,
Register object,
Register index) {
PushSafepointRegistersScope scope(this);
__ Push(object);
__ Push(index);
__ mov(cp, Operand::Zero());
__ CallRuntimeSaveDoubles(Runtime::kLoadMutableDouble);
RecordSafepointWithRegisters(
instr->pointer_map(), 2, Safepoint::kNoLazyDeopt);
__ StoreToSafepointRegisterSlot(r0, result);
}
void LCodeGen::DoLoadFieldByIndex(LLoadFieldByIndex* instr) {
class DeferredLoadMutableDouble FINAL : public LDeferredCode {
public:
DeferredLoadMutableDouble(LCodeGen* codegen,
LLoadFieldByIndex* instr,
Register result,
Register object,
Register index)
: LDeferredCode(codegen),
instr_(instr),
result_(result),
object_(object),
index_(index) {
}
void Generate() OVERRIDE {
codegen()->DoDeferredLoadMutableDouble(instr_, result_, object_, index_);
}
LInstruction* instr() OVERRIDE { return instr_; }
private:
LLoadFieldByIndex* instr_;
Register result_;
Register object_;
Register index_;
};
Register object = ToRegister(instr->object());
Register index = ToRegister(instr->index());
Register result = ToRegister(instr->result());
Register scratch = scratch0();
DeferredLoadMutableDouble* deferred;
deferred = new(zone()) DeferredLoadMutableDouble(
this, instr, result, object, index);
Label out_of_object, done;
__ tst(index, Operand(Smi::FromInt(1)));
__ b(ne, deferred->entry());
__ mov(index, Operand(index, ASR, 1));
__ cmp(index, Operand::Zero());
__ b(lt, &out_of_object);
__ add(scratch, object, Operand::PointerOffsetFromSmiKey(index));
__ ldr(result, FieldMemOperand(scratch, JSObject::kHeaderSize));
__ b(&done);
__ bind(&out_of_object);
__ ldr(result, FieldMemOperand(object, JSObject::kPropertiesOffset));
// Index is equal to negated out of object property index plus 1.
STATIC_ASSERT(kSmiTag == 0 && kSmiTagSize < kPointerSizeLog2);
__ sub(scratch, result, Operand::PointerOffsetFromSmiKey(index));
__ ldr(result, FieldMemOperand(scratch,
FixedArray::kHeaderSize - kPointerSize));
__ bind(deferred->exit());
__ bind(&done);
}
void LCodeGen::DoStoreFrameContext(LStoreFrameContext* instr) {
Register context = ToRegister(instr->context());
__ str(context, MemOperand(fp, StandardFrameConstants::kContextOffset));
}
void LCodeGen::DoAllocateBlockContext(LAllocateBlockContext* instr) {
Handle<ScopeInfo> scope_info = instr->scope_info();
__ Push(scope_info);
__ push(ToRegister(instr->function()));
CallRuntime(Runtime::kPushBlockContext, 2, instr);
RecordSafepoint(Safepoint::kNoLazyDeopt);
}
#undef __
} } // namespace v8::internal