blob: d2534134c700e69cdb1a918a90876e9cf727a391 [file] [log] [blame]
/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
* vim: set ts=8 sts=4 et sw=4 tw=99:
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "jit/shared/CodeGenerator-shared-inl.h"
#include "mozilla/DebugOnly.h"
#include "jit/CompactBuffer.h"
#include "jit/IonCaches.h"
#include "jit/JitcodeMap.h"
#include "jit/JitSpewer.h"
#include "jit/MacroAssembler.h"
#include "jit/MIR.h"
#include "jit/MIRGenerator.h"
#include "jit/OptimizationTracking.h"
#include "js/Conversions.h"
#include "vm/TraceLogging.h"
#include "jit/JitFrames-inl.h"
#include "jit/MacroAssembler-inl.h"
using namespace js;
using namespace js::jit;
using mozilla::BitwiseCast;
using mozilla::DebugOnly;
namespace js {
namespace jit {
MacroAssembler&
CodeGeneratorShared::ensureMasm(MacroAssembler* masmArg)
{
if (masmArg)
return *masmArg;
maybeMasm_.emplace();
return *maybeMasm_;
}
CodeGeneratorShared::CodeGeneratorShared(MIRGenerator* gen, LIRGraph* graph, MacroAssembler* masmArg)
: maybeMasm_(),
masm(ensureMasm(masmArg)),
gen(gen),
graph(*graph),
current(nullptr),
snapshots_(),
recovers_(),
deoptTable_(nullptr),
#ifdef DEBUG
pushedArgs_(0),
#endif
lastOsiPointOffset_(0),
safepoints_(graph->totalSlotCount(), (gen->info().nargs() + 1) * sizeof(Value)),
returnLabel_(),
stubSpace_(),
nativeToBytecodeMap_(nullptr),
nativeToBytecodeMapSize_(0),
nativeToBytecodeTableOffset_(0),
nativeToBytecodeNumRegions_(0),
nativeToBytecodeScriptList_(nullptr),
nativeToBytecodeScriptListLength_(0),
trackedOptimizationsMap_(nullptr),
trackedOptimizationsMapSize_(0),
trackedOptimizationsRegionTableOffset_(0),
trackedOptimizationsTypesTableOffset_(0),
trackedOptimizationsAttemptsTableOffset_(0),
osrEntryOffset_(0),
skipArgCheckEntryOffset_(0),
#ifdef CHECK_OSIPOINT_REGISTERS
checkOsiPointRegisters(JitOptions.checkOsiPointRegisters),
#endif
frameDepth_(graph->paddedLocalSlotsSize() + graph->argumentsSize()),
frameInitialAdjustment_(0)
{
if (gen->isProfilerInstrumentationEnabled())
masm.enableProfilingInstrumentation();
if (gen->compilingAsmJS()) {
// Since asm.js uses the system ABI which does not necessarily use a
// regular array where all slots are sizeof(Value), it maintains the max
// argument stack depth separately.
MOZ_ASSERT(graph->argumentSlotCount() == 0);
frameDepth_ += gen->maxAsmJSStackArgBytes();
if (gen->usesSimd()) {
// If the function uses any SIMD then we may need to insert padding
// so that local slots are aligned for SIMD.
frameInitialAdjustment_ = ComputeByteAlignment(sizeof(AsmJSFrame),
AsmJSStackAlignment);
frameDepth_ += frameInitialAdjustment_;
// Keep the stack aligned. Some SIMD sequences build values on the
// stack and need the stack aligned.
frameDepth_ += ComputeByteAlignment(sizeof(AsmJSFrame) + frameDepth_,
AsmJSStackAlignment);
} else if (gen->performsCall()) {
// An MAsmJSCall does not align the stack pointer at calls sites but
// instead relies on the a priori stack adjustment. This must be the
// last adjustment of frameDepth_.
frameDepth_ += ComputeByteAlignment(sizeof(AsmJSFrame) + frameDepth_,
AsmJSStackAlignment);
}
// FrameSizeClass is only used for bailing, which cannot happen in
// asm.js code.
frameClass_ = FrameSizeClass::None();
} else {
frameClass_ = FrameSizeClass::FromDepth(frameDepth_);
}
}
bool
CodeGeneratorShared::generatePrologue()
{
MOZ_ASSERT(masm.framePushed() == 0);
MOZ_ASSERT(!gen->compilingAsmJS());
#ifdef JS_USE_LINK_REGISTER
masm.pushReturnAddress();
#endif
// If profiling, save the current frame pointer to a per-thread global field.
if (isProfilerInstrumentationEnabled())
masm.profilerEnterFrame(masm.getStackPointer(), CallTempReg0);
// Ensure that the Ion frame is properly aligned.
masm.assertStackAlignment(JitStackAlignment, 0);
// Note that this automatically sets MacroAssembler::framePushed().
masm.reserveStack(frameSize());
masm.checkStackAlignment();
emitTracelogIonStart();
return true;
}
bool
CodeGeneratorShared::generateEpilogue()
{
MOZ_ASSERT(!gen->compilingAsmJS());
masm.bind(&returnLabel_);
emitTracelogIonStop();
masm.freeStack(frameSize());
MOZ_ASSERT(masm.framePushed() == 0);
// If profiling, reset the per-thread global lastJitFrame to point to
// the previous frame.
if (isProfilerInstrumentationEnabled())
masm.profilerExitFrame();
masm.ret();
// On systems that use a constant pool, this is a good time to emit.
masm.flushBuffer();
return true;
}
bool
CodeGeneratorShared::generateOutOfLineCode()
{
for (size_t i = 0; i < outOfLineCode_.length(); i++) {
// Add native => bytecode mapping entries for OOL sites.
// Not enabled on asm.js yet since asm doesn't contain bytecode mappings.
if (!gen->compilingAsmJS()) {
if (!addNativeToBytecodeEntry(outOfLineCode_[i]->bytecodeSite()))
return false;
}
if (!gen->alloc().ensureBallast())
return false;
JitSpew(JitSpew_Codegen, "# Emitting out of line code");
masm.setFramePushed(outOfLineCode_[i]->framePushed());
lastPC_ = outOfLineCode_[i]->pc();
outOfLineCode_[i]->bind(&masm);
outOfLineCode_[i]->generate(this);
}
return !masm.oom();
}
void
CodeGeneratorShared::addOutOfLineCode(OutOfLineCode* code, const MInstruction* mir)
{
MOZ_ASSERT(mir);
addOutOfLineCode(code, mir->trackedSite());
}
void
CodeGeneratorShared::addOutOfLineCode(OutOfLineCode* code, const BytecodeSite* site)
{
code->setFramePushed(masm.framePushed());
code->setBytecodeSite(site);
MOZ_ASSERT_IF(!gen->compilingAsmJS(), code->script()->containsPC(code->pc()));
masm.propagateOOM(outOfLineCode_.append(code));
}
bool
CodeGeneratorShared::addNativeToBytecodeEntry(const BytecodeSite* site)
{
// Skip the table entirely if profiling is not enabled.
if (!isProfilerInstrumentationEnabled())
return true;
// Fails early if the last added instruction caused the macro assembler to
// run out of memory as continuity assumption below do not hold.
if (masm.oom())
return false;
MOZ_ASSERT(site);
MOZ_ASSERT(site->tree());
MOZ_ASSERT(site->pc());
InlineScriptTree* tree = site->tree();
jsbytecode* pc = site->pc();
uint32_t nativeOffset = masm.currentOffset();
MOZ_ASSERT_IF(nativeToBytecodeList_.empty(), nativeOffset == 0);
if (!nativeToBytecodeList_.empty()) {
size_t lastIdx = nativeToBytecodeList_.length() - 1;
NativeToBytecode& lastEntry = nativeToBytecodeList_[lastIdx];
MOZ_ASSERT(nativeOffset >= lastEntry.nativeOffset.offset());
// If the new entry is for the same inlineScriptTree and same
// bytecodeOffset, but the nativeOffset has changed, do nothing.
// The same site just generated some more code.
if (lastEntry.tree == tree && lastEntry.pc == pc) {
JitSpew(JitSpew_Profiling, " => In-place update [%u-%u]",
lastEntry.nativeOffset.offset(), nativeOffset);
return true;
}
// If the new entry is for the same native offset, then update the
// previous entry with the new bytecode site, since the previous
// bytecode site did not generate any native code.
if (lastEntry.nativeOffset.offset() == nativeOffset) {
lastEntry.tree = tree;
lastEntry.pc = pc;
JitSpew(JitSpew_Profiling, " => Overwriting zero-length native region.");
// This overwrite might have made the entry merge-able with a
// previous one. If so, merge it.
if (lastIdx > 0) {
NativeToBytecode& nextToLastEntry = nativeToBytecodeList_[lastIdx - 1];
if (nextToLastEntry.tree == lastEntry.tree && nextToLastEntry.pc == lastEntry.pc) {
JitSpew(JitSpew_Profiling, " => Merging with previous region");
nativeToBytecodeList_.erase(&lastEntry);
}
}
dumpNativeToBytecodeEntry(nativeToBytecodeList_.length() - 1);
return true;
}
}
// Otherwise, some native code was generated for the previous bytecode site.
// Add a new entry for code that is about to be generated.
NativeToBytecode entry;
entry.nativeOffset = CodeOffset(nativeOffset);
entry.tree = tree;
entry.pc = pc;
if (!nativeToBytecodeList_.append(entry))
return false;
JitSpew(JitSpew_Profiling, " => Push new entry.");
dumpNativeToBytecodeEntry(nativeToBytecodeList_.length() - 1);
return true;
}
void
CodeGeneratorShared::dumpNativeToBytecodeEntries()
{
#ifdef JS_JITSPEW
InlineScriptTree* topTree = gen->info().inlineScriptTree();
JitSpewStart(JitSpew_Profiling, "Native To Bytecode Entries for %s:%d\n",
topTree->script()->filename(), topTree->script()->lineno());
for (unsigned i = 0; i < nativeToBytecodeList_.length(); i++)
dumpNativeToBytecodeEntry(i);
#endif
}
void
CodeGeneratorShared::dumpNativeToBytecodeEntry(uint32_t idx)
{
#ifdef JS_JITSPEW
NativeToBytecode& ref = nativeToBytecodeList_[idx];
InlineScriptTree* tree = ref.tree;
JSScript* script = tree->script();
uint32_t nativeOffset = ref.nativeOffset.offset();
unsigned nativeDelta = 0;
unsigned pcDelta = 0;
if (idx + 1 < nativeToBytecodeList_.length()) {
NativeToBytecode* nextRef = &ref + 1;
nativeDelta = nextRef->nativeOffset.offset() - nativeOffset;
if (nextRef->tree == ref.tree)
pcDelta = nextRef->pc - ref.pc;
}
JitSpewStart(JitSpew_Profiling, " %08x [+%-6d] => %-6d [%-4d] {%-10s} (%s:%d",
ref.nativeOffset.offset(),
nativeDelta,
ref.pc - script->code(),
pcDelta,
CodeName[JSOp(*ref.pc)],
script->filename(), script->lineno());
for (tree = tree->caller(); tree; tree = tree->caller()) {
JitSpewCont(JitSpew_Profiling, " <= %s:%d", tree->script()->filename(),
tree->script()->lineno());
}
JitSpewCont(JitSpew_Profiling, ")");
JitSpewFin(JitSpew_Profiling);
#endif
}
bool
CodeGeneratorShared::addTrackedOptimizationsEntry(const TrackedOptimizations* optimizations)
{
if (!isOptimizationTrackingEnabled())
return true;
MOZ_ASSERT(optimizations);
uint32_t nativeOffset = masm.currentOffset();
if (!trackedOptimizations_.empty()) {
NativeToTrackedOptimizations& lastEntry = trackedOptimizations_.back();
MOZ_ASSERT_IF(!masm.oom(), nativeOffset >= lastEntry.endOffset.offset());
// If we're still generating code for the same set of optimizations,
// we are done.
if (lastEntry.optimizations == optimizations)
return true;
}
// If we're generating code for a new set of optimizations, add a new
// entry.
NativeToTrackedOptimizations entry;
entry.startOffset = CodeOffset(nativeOffset);
entry.endOffset = CodeOffset(nativeOffset);
entry.optimizations = optimizations;
return trackedOptimizations_.append(entry);
}
void
CodeGeneratorShared::extendTrackedOptimizationsEntry(const TrackedOptimizations* optimizations)
{
if (!isOptimizationTrackingEnabled())
return;
uint32_t nativeOffset = masm.currentOffset();
NativeToTrackedOptimizations& entry = trackedOptimizations_.back();
MOZ_ASSERT(entry.optimizations == optimizations);
MOZ_ASSERT_IF(!masm.oom(), nativeOffset >= entry.endOffset.offset());
entry.endOffset = CodeOffset(nativeOffset);
// If we generated no code, remove the last entry.
if (nativeOffset == entry.startOffset.offset())
trackedOptimizations_.popBack();
}
// see OffsetOfFrameSlot
static inline int32_t
ToStackIndex(LAllocation* a)
{
if (a->isStackSlot()) {
MOZ_ASSERT(a->toStackSlot()->slot() >= 1);
return a->toStackSlot()->slot();
}
return -int32_t(sizeof(JitFrameLayout) + a->toArgument()->index());
}
void
CodeGeneratorShared::encodeAllocation(LSnapshot* snapshot, MDefinition* mir,
uint32_t* allocIndex)
{
if (mir->isBox())
mir = mir->toBox()->getOperand(0);
MIRType type =
mir->isRecoveredOnBailout() ? MIRType_None :
mir->isUnused() ? MIRType_MagicOptimizedOut :
mir->type();
RValueAllocation alloc;
switch (type) {
case MIRType_None:
{
MOZ_ASSERT(mir->isRecoveredOnBailout());
uint32_t index = 0;
LRecoverInfo* recoverInfo = snapshot->recoverInfo();
MNode** it = recoverInfo->begin();
MNode** end = recoverInfo->end();
while (it != end && mir != *it) {
++it;
++index;
}
// This MDefinition is recovered, thus it should be listed in the
// LRecoverInfo.
MOZ_ASSERT(it != end && mir == *it);
// Lambda should have a default value readable for iterating over the
// inner frames.
if (mir->isLambda()) {
MConstant* constant = mir->toLambda()->functionOperand();
uint32_t cstIndex;
masm.propagateOOM(graph.addConstantToPool(constant->value(), &cstIndex));
alloc = RValueAllocation::RecoverInstruction(index, cstIndex);
break;
}
alloc = RValueAllocation::RecoverInstruction(index);
break;
}
case MIRType_Undefined:
alloc = RValueAllocation::Undefined();
break;
case MIRType_Null:
alloc = RValueAllocation::Null();
break;
case MIRType_Int32:
case MIRType_String:
case MIRType_Symbol:
case MIRType_Object:
case MIRType_ObjectOrNull:
case MIRType_Boolean:
case MIRType_Double:
{
LAllocation* payload = snapshot->payloadOfSlot(*allocIndex);
if (payload->isConstant()) {
MConstant* constant = mir->toConstant();
uint32_t index;
masm.propagateOOM(graph.addConstantToPool(constant->value(), &index));
alloc = RValueAllocation::ConstantPool(index);
break;
}
JSValueType valueType =
(type == MIRType_ObjectOrNull) ? JSVAL_TYPE_OBJECT : ValueTypeFromMIRType(type);
MOZ_ASSERT(payload->isMemory() || payload->isRegister());
if (payload->isMemory())
alloc = RValueAllocation::Typed(valueType, ToStackIndex(payload));
else if (payload->isGeneralReg())
alloc = RValueAllocation::Typed(valueType, ToRegister(payload));
else if (payload->isFloatReg())
alloc = RValueAllocation::Double(ToFloatRegister(payload));
break;
}
case MIRType_Float32:
case MIRType_Int32x4:
case MIRType_Float32x4:
{
LAllocation* payload = snapshot->payloadOfSlot(*allocIndex);
if (payload->isConstant()) {
MConstant* constant = mir->toConstant();
uint32_t index;
masm.propagateOOM(graph.addConstantToPool(constant->value(), &index));
alloc = RValueAllocation::ConstantPool(index);
break;
}
MOZ_ASSERT(payload->isMemory() || payload->isFloatReg());
if (payload->isFloatReg())
alloc = RValueAllocation::AnyFloat(ToFloatRegister(payload));
else
alloc = RValueAllocation::AnyFloat(ToStackIndex(payload));
break;
}
case MIRType_MagicOptimizedArguments:
case MIRType_MagicOptimizedOut:
case MIRType_MagicUninitializedLexical:
{
uint32_t index;
Value v = MagicValue(type == MIRType_MagicOptimizedArguments
? JS_OPTIMIZED_ARGUMENTS
: (type == MIRType_MagicOptimizedOut
? JS_OPTIMIZED_OUT
: JS_UNINITIALIZED_LEXICAL));
masm.propagateOOM(graph.addConstantToPool(v, &index));
alloc = RValueAllocation::ConstantPool(index);
break;
}
default:
{
MOZ_ASSERT(mir->type() == MIRType_Value);
LAllocation* payload = snapshot->payloadOfSlot(*allocIndex);
#ifdef JS_NUNBOX32
LAllocation* type = snapshot->typeOfSlot(*allocIndex);
if (type->isRegister()) {
if (payload->isRegister())
alloc = RValueAllocation::Untyped(ToRegister(type), ToRegister(payload));
else
alloc = RValueAllocation::Untyped(ToRegister(type), ToStackIndex(payload));
} else {
if (payload->isRegister())
alloc = RValueAllocation::Untyped(ToStackIndex(type), ToRegister(payload));
else
alloc = RValueAllocation::Untyped(ToStackIndex(type), ToStackIndex(payload));
}
#elif JS_PUNBOX64
if (payload->isRegister())
alloc = RValueAllocation::Untyped(ToRegister(payload));
else
alloc = RValueAllocation::Untyped(ToStackIndex(payload));
#endif
break;
}
}
// This set an extra bit as part of the RValueAllocation, such that we know
// that recover instruction have to be executed without wrapping the
// instruction in a no-op recover instruction.
if (mir->isIncompleteObject())
alloc.setNeedSideEffect();
snapshots_.add(alloc);
*allocIndex += mir->isRecoveredOnBailout() ? 0 : 1;
}
void
CodeGeneratorShared::encode(LRecoverInfo* recover)
{
if (recover->recoverOffset() != INVALID_RECOVER_OFFSET)
return;
uint32_t numInstructions = recover->numInstructions();
JitSpew(JitSpew_IonSnapshots, "Encoding LRecoverInfo %p (frameCount %u, instructions %u)",
(void*)recover, recover->mir()->frameCount(), numInstructions);
MResumePoint::Mode mode = recover->mir()->mode();
MOZ_ASSERT(mode != MResumePoint::Outer);
bool resumeAfter = (mode == MResumePoint::ResumeAfter);
RecoverOffset offset = recovers_.startRecover(numInstructions, resumeAfter);
for (MNode* insn : *recover)
recovers_.writeInstruction(insn);
recovers_.endRecover();
recover->setRecoverOffset(offset);
masm.propagateOOM(!recovers_.oom());
}
void
CodeGeneratorShared::encode(LSnapshot* snapshot)
{
if (snapshot->snapshotOffset() != INVALID_SNAPSHOT_OFFSET)
return;
LRecoverInfo* recoverInfo = snapshot->recoverInfo();
encode(recoverInfo);
RecoverOffset recoverOffset = recoverInfo->recoverOffset();
MOZ_ASSERT(recoverOffset != INVALID_RECOVER_OFFSET);
JitSpew(JitSpew_IonSnapshots, "Encoding LSnapshot %p (LRecover %p)",
(void*)snapshot, (void*) recoverInfo);
SnapshotOffset offset = snapshots_.startSnapshot(recoverOffset, snapshot->bailoutKind());
#ifdef TRACK_SNAPSHOTS
uint32_t pcOpcode = 0;
uint32_t lirOpcode = 0;
uint32_t lirId = 0;
uint32_t mirOpcode = 0;
uint32_t mirId = 0;
if (LNode* ins = instruction()) {
lirOpcode = ins->op();
lirId = ins->id();
if (ins->mirRaw()) {
mirOpcode = ins->mirRaw()->op();
mirId = ins->mirRaw()->id();
if (ins->mirRaw()->trackedPc())
pcOpcode = *ins->mirRaw()->trackedPc();
}
}
snapshots_.trackSnapshot(pcOpcode, mirOpcode, mirId, lirOpcode, lirId);
#endif
uint32_t allocIndex = 0;
for (LRecoverInfo::OperandIter it(recoverInfo); !it; ++it) {
DebugOnly<uint32_t> allocWritten = snapshots_.allocWritten();
encodeAllocation(snapshot, *it, &allocIndex);
MOZ_ASSERT_IF(!snapshots_.oom(), allocWritten + 1 == snapshots_.allocWritten());
}
MOZ_ASSERT(allocIndex == snapshot->numSlots());
snapshots_.endSnapshot();
snapshot->setSnapshotOffset(offset);
masm.propagateOOM(!snapshots_.oom());
}
bool
CodeGeneratorShared::assignBailoutId(LSnapshot* snapshot)
{
MOZ_ASSERT(snapshot->snapshotOffset() != INVALID_SNAPSHOT_OFFSET);
// Can we not use bailout tables at all?
if (!deoptTable_)
return false;
MOZ_ASSERT(frameClass_ != FrameSizeClass::None());
if (snapshot->bailoutId() != INVALID_BAILOUT_ID)
return true;
// Is the bailout table full?
if (bailouts_.length() >= BAILOUT_TABLE_SIZE)
return false;
unsigned bailoutId = bailouts_.length();
snapshot->setBailoutId(bailoutId);
JitSpew(JitSpew_IonSnapshots, "Assigned snapshot bailout id %u", bailoutId);
return bailouts_.append(snapshot->snapshotOffset());
}
bool
CodeGeneratorShared::encodeSafepoints()
{
for (SafepointIndex& index : safepointIndices_) {
LSafepoint* safepoint = index.safepoint();
if (!safepoint->encoded())
safepoints_.encode(safepoint);
index.resolve();
}
return !safepoints_.oom();
}
bool
CodeGeneratorShared::createNativeToBytecodeScriptList(JSContext* cx)
{
js::Vector<JSScript*, 0, SystemAllocPolicy> scriptList;
InlineScriptTree* tree = gen->info().inlineScriptTree();
for (;;) {
// Add script from current tree.
bool found = false;
for (uint32_t i = 0; i < scriptList.length(); i++) {
if (scriptList[i] == tree->script()) {
found = true;
break;
}
}
if (!found) {
if (!scriptList.append(tree->script()))
return false;
}
// Process rest of tree
// If children exist, emit children.
if (tree->hasChildren()) {
tree = tree->firstChild();
continue;
}
// Otherwise, find the first tree up the chain (including this one)
// that contains a next sibling.
while (!tree->hasNextCallee() && tree->hasCaller())
tree = tree->caller();
// If we found a sibling, use it.
if (tree->hasNextCallee()) {
tree = tree->nextCallee();
continue;
}
// Otherwise, we must have reached the top without finding any siblings.
MOZ_ASSERT(tree->isOutermostCaller());
break;
}
// Allocate array for list.
JSScript** data = cx->runtime()->pod_malloc<JSScript*>(scriptList.length());
if (!data)
return false;
for (uint32_t i = 0; i < scriptList.length(); i++)
data[i] = scriptList[i];
// Success.
nativeToBytecodeScriptListLength_ = scriptList.length();
nativeToBytecodeScriptList_ = data;
return true;
}
bool
CodeGeneratorShared::generateCompactNativeToBytecodeMap(JSContext* cx, JitCode* code)
{
MOZ_ASSERT(nativeToBytecodeScriptListLength_ == 0);
MOZ_ASSERT(nativeToBytecodeScriptList_ == nullptr);
MOZ_ASSERT(nativeToBytecodeMap_ == nullptr);
MOZ_ASSERT(nativeToBytecodeMapSize_ == 0);
MOZ_ASSERT(nativeToBytecodeTableOffset_ == 0);
MOZ_ASSERT(nativeToBytecodeNumRegions_ == 0);
if (!createNativeToBytecodeScriptList(cx))
return false;
MOZ_ASSERT(nativeToBytecodeScriptListLength_ > 0);
MOZ_ASSERT(nativeToBytecodeScriptList_ != nullptr);
CompactBufferWriter writer;
uint32_t tableOffset = 0;
uint32_t numRegions = 0;
if (!JitcodeIonTable::WriteIonTable(
writer, nativeToBytecodeScriptList_, nativeToBytecodeScriptListLength_,
&nativeToBytecodeList_[0],
&nativeToBytecodeList_[0] + nativeToBytecodeList_.length(),
&tableOffset, &numRegions))
{
js_free(nativeToBytecodeScriptList_);
return false;
}
MOZ_ASSERT(tableOffset > 0);
MOZ_ASSERT(numRegions > 0);
// Writer is done, copy it to sized buffer.
uint8_t* data = cx->runtime()->pod_malloc<uint8_t>(writer.length());
if (!data) {
js_free(nativeToBytecodeScriptList_);
return false;
}
memcpy(data, writer.buffer(), writer.length());
nativeToBytecodeMap_ = data;
nativeToBytecodeMapSize_ = writer.length();
nativeToBytecodeTableOffset_ = tableOffset;
nativeToBytecodeNumRegions_ = numRegions;
verifyCompactNativeToBytecodeMap(code);
JitSpew(JitSpew_Profiling, "Compact Native To Bytecode Map [%p-%p]",
data, data + nativeToBytecodeMapSize_);
return true;
}
void
CodeGeneratorShared::verifyCompactNativeToBytecodeMap(JitCode* code)
{
#ifdef DEBUG
MOZ_ASSERT(nativeToBytecodeScriptListLength_ > 0);
MOZ_ASSERT(nativeToBytecodeScriptList_ != nullptr);
MOZ_ASSERT(nativeToBytecodeMap_ != nullptr);
MOZ_ASSERT(nativeToBytecodeMapSize_ > 0);
MOZ_ASSERT(nativeToBytecodeTableOffset_ > 0);
MOZ_ASSERT(nativeToBytecodeNumRegions_ > 0);
// The pointer to the table must be 4-byte aligned
const uint8_t* tablePtr = nativeToBytecodeMap_ + nativeToBytecodeTableOffset_;
MOZ_ASSERT(uintptr_t(tablePtr) % sizeof(uint32_t) == 0);
// Verify that numRegions was encoded correctly.
const JitcodeIonTable* ionTable = reinterpret_cast<const JitcodeIonTable*>(tablePtr);
MOZ_ASSERT(ionTable->numRegions() == nativeToBytecodeNumRegions_);
// Region offset for first region should be at the start of the payload region.
// Since the offsets are backward from the start of the table, the first entry
// backoffset should be equal to the forward table offset from the start of the
// allocated data.
MOZ_ASSERT(ionTable->regionOffset(0) == nativeToBytecodeTableOffset_);
// Verify each region.
for (uint32_t i = 0; i < ionTable->numRegions(); i++) {
// Back-offset must point into the payload region preceding the table, not before it.
MOZ_ASSERT(ionTable->regionOffset(i) <= nativeToBytecodeTableOffset_);
// Back-offset must point to a later area in the payload region than previous
// back-offset. This means that back-offsets decrease monotonically.
MOZ_ASSERT_IF(i > 0, ionTable->regionOffset(i) < ionTable->regionOffset(i - 1));
JitcodeRegionEntry entry = ionTable->regionEntry(i);
// Ensure native code offset for region falls within jitcode.
MOZ_ASSERT(entry.nativeOffset() <= code->instructionsSize());
// Read out script/pc stack and verify.
JitcodeRegionEntry::ScriptPcIterator scriptPcIter = entry.scriptPcIterator();
while (scriptPcIter.hasMore()) {
uint32_t scriptIdx = 0, pcOffset = 0;
scriptPcIter.readNext(&scriptIdx, &pcOffset);
// Ensure scriptIdx refers to a valid script in the list.
MOZ_ASSERT(scriptIdx < nativeToBytecodeScriptListLength_);
JSScript* script = nativeToBytecodeScriptList_[scriptIdx];
// Ensure pcOffset falls within the script.
MOZ_ASSERT(pcOffset < script->length());
}
// Obtain the original nativeOffset and pcOffset and script.
uint32_t curNativeOffset = entry.nativeOffset();
JSScript* script = nullptr;
uint32_t curPcOffset = 0;
{
uint32_t scriptIdx = 0;
scriptPcIter.reset();
scriptPcIter.readNext(&scriptIdx, &curPcOffset);
script = nativeToBytecodeScriptList_[scriptIdx];
}
// Read out nativeDeltas and pcDeltas and verify.
JitcodeRegionEntry::DeltaIterator deltaIter = entry.deltaIterator();
while (deltaIter.hasMore()) {
uint32_t nativeDelta = 0;
int32_t pcDelta = 0;
deltaIter.readNext(&nativeDelta, &pcDelta);
curNativeOffset += nativeDelta;
curPcOffset = uint32_t(int32_t(curPcOffset) + pcDelta);
// Ensure that nativeOffset still falls within jitcode after delta.
MOZ_ASSERT(curNativeOffset <= code->instructionsSize());
// Ensure that pcOffset still falls within bytecode after delta.
MOZ_ASSERT(curPcOffset < script->length());
}
}
#endif // DEBUG
}
bool
CodeGeneratorShared::generateCompactTrackedOptimizationsMap(JSContext* cx, JitCode* code,
IonTrackedTypeVector* allTypes)
{
MOZ_ASSERT(trackedOptimizationsMap_ == nullptr);
MOZ_ASSERT(trackedOptimizationsMapSize_ == 0);
MOZ_ASSERT(trackedOptimizationsRegionTableOffset_ == 0);
MOZ_ASSERT(trackedOptimizationsTypesTableOffset_ == 0);
MOZ_ASSERT(trackedOptimizationsAttemptsTableOffset_ == 0);
if (trackedOptimizations_.empty())
return true;
UniqueTrackedOptimizations unique(cx);
if (!unique.init())
return false;
// Iterate through all entries to deduplicate their optimization attempts.
for (size_t i = 0; i < trackedOptimizations_.length(); i++) {
NativeToTrackedOptimizations& entry = trackedOptimizations_[i];
if (!unique.add(entry.optimizations))
return false;
}
// Sort the unique optimization attempts by frequency to stabilize the
// attempts' indices in the compact table we will write later.
if (!unique.sortByFrequency(cx))
return false;
// Write out the ranges and the table.
CompactBufferWriter writer;
uint32_t numRegions;
uint32_t regionTableOffset;
uint32_t typesTableOffset;
uint32_t attemptsTableOffset;
if (!WriteIonTrackedOptimizationsTable(cx, writer,
trackedOptimizations_.begin(),
trackedOptimizations_.end(),
unique, &numRegions,
&regionTableOffset, &typesTableOffset,
&attemptsTableOffset, allTypes))
{
return false;
}
MOZ_ASSERT(regionTableOffset > 0);
MOZ_ASSERT(typesTableOffset > 0);
MOZ_ASSERT(attemptsTableOffset > 0);
MOZ_ASSERT(typesTableOffset > regionTableOffset);
MOZ_ASSERT(attemptsTableOffset > typesTableOffset);
// Copy over the table out of the writer's buffer.
uint8_t* data = cx->runtime()->pod_malloc<uint8_t>(writer.length());
if (!data)
return false;
memcpy(data, writer.buffer(), writer.length());
trackedOptimizationsMap_ = data;
trackedOptimizationsMapSize_ = writer.length();
trackedOptimizationsRegionTableOffset_ = regionTableOffset;
trackedOptimizationsTypesTableOffset_ = typesTableOffset;
trackedOptimizationsAttemptsTableOffset_ = attemptsTableOffset;
verifyCompactTrackedOptimizationsMap(code, numRegions, unique, allTypes);
JitSpew(JitSpew_OptimizationTracking,
"== Compact Native To Optimizations Map [%p-%p] size %u",
data, data + trackedOptimizationsMapSize_, trackedOptimizationsMapSize_);
JitSpew(JitSpew_OptimizationTracking,
" with type list of length %u, size %u",
allTypes->length(), allTypes->length() * sizeof(IonTrackedTypeWithAddendum));
return true;
}
#ifdef DEBUG
class ReadTempAttemptsVectorOp : public JS::ForEachTrackedOptimizationAttemptOp
{
TempOptimizationAttemptsVector* attempts_;
bool oom_;
public:
explicit ReadTempAttemptsVectorOp(TempOptimizationAttemptsVector* attempts)
: attempts_(attempts), oom_(false)
{ }
bool oom() {
return oom_;
}
void operator()(JS::TrackedStrategy strategy, JS::TrackedOutcome outcome) override {
if (!attempts_->append(OptimizationAttempt(strategy, outcome)))
oom_ = true;
}
};
struct ReadTempTypeInfoVectorOp : public IonTrackedOptimizationsTypeInfo::ForEachOp
{
TempAllocator& alloc_;
TempOptimizationTypeInfoVector* types_;
TempTypeList accTypes_;
bool oom_;
public:
ReadTempTypeInfoVectorOp(TempAllocator& alloc, TempOptimizationTypeInfoVector* types)
: alloc_(alloc),
types_(types),
accTypes_(alloc),
oom_(false)
{ }
bool oom() {
return oom_;
}
void readType(const IonTrackedTypeWithAddendum& tracked) override {
if (!accTypes_.append(tracked.type))
oom_ = true;
}
void operator()(JS::TrackedTypeSite site, MIRType mirType) override {
OptimizationTypeInfo ty(alloc_, site, mirType);
for (uint32_t i = 0; i < accTypes_.length(); i++) {
if (!ty.trackType(accTypes_[i]))
oom_ = true;
}
if (!types_->append(mozilla::Move(ty)))
oom_ = true;
accTypes_.clear();
}
};
#endif // DEBUG
void
CodeGeneratorShared::verifyCompactTrackedOptimizationsMap(JitCode* code, uint32_t numRegions,
const UniqueTrackedOptimizations& unique,
const IonTrackedTypeVector* allTypes)
{
#ifdef DEBUG
MOZ_ASSERT(trackedOptimizationsMap_ != nullptr);
MOZ_ASSERT(trackedOptimizationsMapSize_ > 0);
MOZ_ASSERT(trackedOptimizationsRegionTableOffset_ > 0);
MOZ_ASSERT(trackedOptimizationsTypesTableOffset_ > 0);
MOZ_ASSERT(trackedOptimizationsAttemptsTableOffset_ > 0);
// Table pointers must all be 4-byte aligned.
const uint8_t* regionTableAddr = trackedOptimizationsMap_ +
trackedOptimizationsRegionTableOffset_;
const uint8_t* typesTableAddr = trackedOptimizationsMap_ +
trackedOptimizationsTypesTableOffset_;
const uint8_t* attemptsTableAddr = trackedOptimizationsMap_ +
trackedOptimizationsAttemptsTableOffset_;
MOZ_ASSERT(uintptr_t(regionTableAddr) % sizeof(uint32_t) == 0);
MOZ_ASSERT(uintptr_t(typesTableAddr) % sizeof(uint32_t) == 0);
MOZ_ASSERT(uintptr_t(attemptsTableAddr) % sizeof(uint32_t) == 0);
// Assert that the number of entries matches up for the tables.
const IonTrackedOptimizationsRegionTable* regionTable =
(const IonTrackedOptimizationsRegionTable*) regionTableAddr;
MOZ_ASSERT(regionTable->numEntries() == numRegions);
const IonTrackedOptimizationsTypesTable* typesTable =
(const IonTrackedOptimizationsTypesTable*) typesTableAddr;
MOZ_ASSERT(typesTable->numEntries() == unique.count());
const IonTrackedOptimizationsAttemptsTable* attemptsTable =
(const IonTrackedOptimizationsAttemptsTable*) attemptsTableAddr;
MOZ_ASSERT(attemptsTable->numEntries() == unique.count());
// Verify each region.
uint32_t trackedIdx = 0;
for (uint32_t regionIdx = 0; regionIdx < regionTable->numEntries(); regionIdx++) {
// Check reverse offsets are within bounds.
MOZ_ASSERT(regionTable->entryOffset(regionIdx) <= trackedOptimizationsRegionTableOffset_);
MOZ_ASSERT_IF(regionIdx > 0, regionTable->entryOffset(regionIdx) <
regionTable->entryOffset(regionIdx - 1));
IonTrackedOptimizationsRegion region = regionTable->entry(regionIdx);
// Check the region range is covered by jitcode.
MOZ_ASSERT(region.startOffset() <= code->instructionsSize());
MOZ_ASSERT(region.endOffset() <= code->instructionsSize());
IonTrackedOptimizationsRegion::RangeIterator iter = region.ranges();
while (iter.more()) {
// Assert that the offsets are correctly decoded from the delta.
uint32_t startOffset, endOffset;
uint8_t index;
iter.readNext(&startOffset, &endOffset, &index);
NativeToTrackedOptimizations& entry = trackedOptimizations_[trackedIdx++];
MOZ_ASSERT(startOffset == entry.startOffset.offset());
MOZ_ASSERT(endOffset == entry.endOffset.offset());
MOZ_ASSERT(index == unique.indexOf(entry.optimizations));
// Assert that the type info and attempts vectors are correctly
// decoded. This is disabled for now if the types table might
// contain nursery pointers, in which case the types might not
// match, see bug 1175761.
if (!code->runtimeFromMainThread()->gc.storeBuffer.cancelIonCompilations()) {
IonTrackedOptimizationsTypeInfo typeInfo = typesTable->entry(index);
TempOptimizationTypeInfoVector tvec(alloc());
ReadTempTypeInfoVectorOp top(alloc(), &tvec);
typeInfo.forEach(top, allTypes);
MOZ_ASSERT_IF(!top.oom(), entry.optimizations->matchTypes(tvec));
}
IonTrackedOptimizationsAttempts attempts = attemptsTable->entry(index);
TempOptimizationAttemptsVector avec(alloc());
ReadTempAttemptsVectorOp aop(&avec);
attempts.forEach(aop);
MOZ_ASSERT_IF(!aop.oom(), entry.optimizations->matchAttempts(avec));
}
}
#endif
}
void
CodeGeneratorShared::markSafepoint(LInstruction* ins)
{
markSafepointAt(masm.currentOffset(), ins);
}
void
CodeGeneratorShared::markSafepointAt(uint32_t offset, LInstruction* ins)
{
MOZ_ASSERT_IF(!safepointIndices_.empty() && !masm.oom(),
offset - safepointIndices_.back().displacement() >= sizeof(uint32_t));
masm.propagateOOM(safepointIndices_.append(SafepointIndex(offset, ins->safepoint())));
}
void
CodeGeneratorShared::ensureOsiSpace()
{
// For a refresher, an invalidation point is of the form:
// 1: call <target>
// 2: ...
// 3: <osipoint>
//
// The four bytes *before* instruction 2 are overwritten with an offset.
// Callers must ensure that the instruction itself has enough bytes to
// support this.
//
// The bytes *at* instruction 3 are overwritten with an invalidation jump.
// jump. These bytes may be in a completely different IR sequence, but
// represent the join point of the call out of the function.
//
// At points where we want to ensure that invalidation won't corrupt an
// important instruction, we make sure to pad with nops.
if (masm.currentOffset() - lastOsiPointOffset_ < Assembler::PatchWrite_NearCallSize()) {
int32_t paddingSize = Assembler::PatchWrite_NearCallSize();
paddingSize -= masm.currentOffset() - lastOsiPointOffset_;
for (int32_t i = 0; i < paddingSize; ++i)
masm.nop();
}
MOZ_ASSERT_IF(!masm.oom(),
masm.currentOffset() - lastOsiPointOffset_ >= Assembler::PatchWrite_NearCallSize());
lastOsiPointOffset_ = masm.currentOffset();
}
uint32_t
CodeGeneratorShared::markOsiPoint(LOsiPoint* ins)
{
encode(ins->snapshot());
ensureOsiSpace();
uint32_t offset = masm.currentOffset();
SnapshotOffset so = ins->snapshot()->snapshotOffset();
masm.propagateOOM(osiIndices_.append(OsiIndex(offset, so)));
return offset;
}
#ifdef CHECK_OSIPOINT_REGISTERS
template <class Op>
static void
HandleRegisterDump(Op op, MacroAssembler& masm, LiveRegisterSet liveRegs, Register activation,
Register scratch)
{
const size_t baseOffset = JitActivation::offsetOfRegs();
// Handle live GPRs.
for (GeneralRegisterIterator iter(liveRegs.gprs()); iter.more(); iter++) {
Register reg = *iter;
Address dump(activation, baseOffset + RegisterDump::offsetOfRegister(reg));
if (reg == activation) {
// To use the original value of the activation register (that's
// now on top of the stack), we need the scratch register.
masm.push(scratch);
masm.loadPtr(Address(masm.getStackPointer(), sizeof(uintptr_t)), scratch);
op(scratch, dump);
masm.pop(scratch);
} else {
op(reg, dump);
}
}
// Handle live FPRs.
for (FloatRegisterIterator iter(liveRegs.fpus()); iter.more(); iter++) {
FloatRegister reg = *iter;
Address dump(activation, baseOffset + RegisterDump::offsetOfRegister(reg));
op(reg, dump);
}
}
class StoreOp
{
MacroAssembler& masm;
public:
explicit StoreOp(MacroAssembler& masm)
: masm(masm)
{}
void operator()(Register reg, Address dump) {
masm.storePtr(reg, dump);
}
void operator()(FloatRegister reg, Address dump) {
if (reg.isDouble())
masm.storeDouble(reg, dump);
else if (reg.isSingle())
masm.storeFloat32(reg, dump);
#if defined(JS_CODEGEN_X86) || defined(JS_CODEGEN_X64)
else if (reg.isSimd128())
masm.storeUnalignedFloat32x4(reg, dump);
#endif
else
MOZ_CRASH("Unexpected register type.");
}
};
static void
StoreAllLiveRegs(MacroAssembler& masm, LiveRegisterSet liveRegs)
{
// Store a copy of all live registers before performing the call.
// When we reach the OsiPoint, we can use this to check nothing
// modified them in the meantime.
// Load pointer to the JitActivation in a scratch register.
AllocatableGeneralRegisterSet allRegs(GeneralRegisterSet::All());
Register scratch = allRegs.takeAny();
masm.push(scratch);
masm.loadJitActivation(scratch);
Address checkRegs(scratch, JitActivation::offsetOfCheckRegs());
masm.add32(Imm32(1), checkRegs);
StoreOp op(masm);
HandleRegisterDump<StoreOp>(op, masm, liveRegs, scratch, allRegs.getAny());
masm.pop(scratch);
}
class VerifyOp
{
MacroAssembler& masm;
Label* failure_;
public:
VerifyOp(MacroAssembler& masm, Label* failure)
: masm(masm), failure_(failure)
{}
void operator()(Register reg, Address dump) {
masm.branchPtr(Assembler::NotEqual, dump, reg, failure_);
}
void operator()(FloatRegister reg, Address dump) {
FloatRegister scratch;
if (reg.isDouble()) {
scratch = ScratchDoubleReg;
masm.loadDouble(dump, scratch);
masm.branchDouble(Assembler::DoubleNotEqual, scratch, reg, failure_);
} else if (reg.isSingle()) {
scratch = ScratchFloat32Reg;
masm.loadFloat32(dump, scratch);
masm.branchFloat(Assembler::DoubleNotEqual, scratch, reg, failure_);
}
// :TODO: (Bug 1133745) Add support to verify SIMD registers.
}
};
void
CodeGeneratorShared::verifyOsiPointRegs(LSafepoint* safepoint)
{
// Ensure the live registers stored by callVM did not change between
// the call and this OsiPoint. Try-catch relies on this invariant.
// Load pointer to the JitActivation in a scratch register.
AllocatableGeneralRegisterSet allRegs(GeneralRegisterSet::All());
Register scratch = allRegs.takeAny();
masm.push(scratch);
masm.loadJitActivation(scratch);
// If we should not check registers (because the instruction did not call
// into the VM, or a GC happened), we're done.
Label failure, done;
Address checkRegs(scratch, JitActivation::offsetOfCheckRegs());
masm.branch32(Assembler::Equal, checkRegs, Imm32(0), &done);
// Having more than one VM function call made in one visit function at
// runtime is a sec-ciritcal error, because if we conservatively assume that
// one of the function call can re-enter Ion, then the invalidation process
// will potentially add a call at a random location, by patching the code
// before the return address.
masm.branch32(Assembler::NotEqual, checkRegs, Imm32(1), &failure);
// Set checkRegs to 0, so that we don't try to verify registers after we
// return from this script to the caller.
masm.store32(Imm32(0), checkRegs);
// Ignore clobbered registers. Some instructions (like LValueToInt32) modify
// temps after calling into the VM. This is fine because no other
// instructions (including this OsiPoint) will depend on them. Also
// backtracking can also use the same register for an input and an output.
// These are marked as clobbered and shouldn't get checked.
LiveRegisterSet liveRegs;
liveRegs.set() = RegisterSet::Intersect(safepoint->liveRegs().set(),
RegisterSet::Not(safepoint->clobberedRegs().set()));
VerifyOp op(masm, &failure);
HandleRegisterDump<VerifyOp>(op, masm, liveRegs, scratch, allRegs.getAny());
masm.jump(&done);
// Do not profile the callWithABI that occurs below. This is to avoid a
// rare corner case that occurs when profiling interacts with itself:
//
// When slow profiling assertions are turned on, FunctionBoundary ops
// (which update the profiler pseudo-stack) may emit a callVM, which
// forces them to have an osi point associated with them. The
// FunctionBoundary for inline function entry is added to the caller's
// graph with a PC from the caller's code, but during codegen it modifies
// SPS instrumentation to add the callee as the current top-most script.
// When codegen gets to the OSIPoint, and the callWithABI below is
// emitted, the codegen thinks that the current frame is the callee, but
// the PC it's using from the OSIPoint refers to the caller. This causes
// the profiler instrumentation of the callWithABI below to ASSERT, since
// the script and pc are mismatched. To avoid this, we simply omit
// instrumentation for these callWithABIs.
// Any live register captured by a safepoint (other than temp registers)
// must remain unchanged between the call and the OsiPoint instruction.
masm.bind(&failure);
masm.assumeUnreachable("Modified registers between VM call and OsiPoint");
masm.bind(&done);
masm.pop(scratch);
}
bool
CodeGeneratorShared::shouldVerifyOsiPointRegs(LSafepoint* safepoint)
{
if (!checkOsiPointRegisters)
return false;
if (safepoint->liveRegs().emptyGeneral() && safepoint->liveRegs().emptyFloat())
return false; // No registers to check.
return true;
}
void
CodeGeneratorShared::resetOsiPointRegs(LSafepoint* safepoint)
{
if (!shouldVerifyOsiPointRegs(safepoint))
return;
// Set checkRegs to 0. If we perform a VM call, the instruction
// will set it to 1.
AllocatableGeneralRegisterSet allRegs(GeneralRegisterSet::All());
Register scratch = allRegs.takeAny();
masm.push(scratch);
masm.loadJitActivation(scratch);
Address checkRegs(scratch, JitActivation::offsetOfCheckRegs());
masm.store32(Imm32(0), checkRegs);
masm.pop(scratch);
}
#endif
// Before doing any call to Cpp, you should ensure that volatile
// registers are evicted by the register allocator.
void
CodeGeneratorShared::callVM(const VMFunction& fun, LInstruction* ins, const Register* dynStack)
{
// If we're calling a function with an out parameter type of double, make
// sure we have an FPU.
MOZ_ASSERT_IF(fun.outParam == Type_Double, GetJitContext()->runtime->jitSupportsFloatingPoint());
#ifdef DEBUG
if (ins->mirRaw()) {
MOZ_ASSERT(ins->mirRaw()->isInstruction());
MInstruction* mir = ins->mirRaw()->toInstruction();
MOZ_ASSERT_IF(mir->needsResumePoint(), mir->resumePoint());
}
#endif
#ifdef JS_TRACE_LOGGING
emitTracelogStartEvent(TraceLogger_VM);
#endif
// Stack is:
// ... frame ...
// [args]
#ifdef DEBUG
MOZ_ASSERT(pushedArgs_ == fun.explicitArgs);
pushedArgs_ = 0;
#endif
// Get the wrapper of the VM function.
JitCode* wrapper = gen->jitRuntime()->getVMWrapper(fun);
if (!wrapper) {
masm.setOOM();
return;
}
#ifdef CHECK_OSIPOINT_REGISTERS
if (shouldVerifyOsiPointRegs(ins->safepoint()))
StoreAllLiveRegs(masm, ins->safepoint()->liveRegs());
#endif
// Push an exit frame descriptor. If |dynStack| is a valid pointer to a
// register, then its value is added to the value of the |framePushed()| to
// fill the frame descriptor.
if (dynStack) {
masm.addPtr(Imm32(masm.framePushed()), *dynStack);
masm.makeFrameDescriptor(*dynStack, JitFrame_IonJS);
masm.Push(*dynStack); // descriptor
} else {
masm.pushStaticFrameDescriptor(JitFrame_IonJS);
}
// Call the wrapper function. The wrapper is in charge to unwind the stack
// when returning from the call. Failures are handled with exceptions based
// on the return value of the C functions. To guard the outcome of the
// returned value, use another LIR instruction.
uint32_t callOffset = masm.callJit(wrapper);
markSafepointAt(callOffset, ins);
// Remove rest of the frame left on the stack. We remove the return address
// which is implicitly poped when returning.
int framePop = sizeof(ExitFrameLayout) - sizeof(void*);
// Pop arguments from framePushed.
masm.implicitPop(fun.explicitStackSlots() * sizeof(void*) + framePop);
// Stack is:
// ... frame ...
#ifdef JS_TRACE_LOGGING
emitTracelogStopEvent(TraceLogger_VM);
#endif
}
class OutOfLineTruncateSlow : public OutOfLineCodeBase<CodeGeneratorShared>
{
FloatRegister src_;
Register dest_;
bool needFloat32Conversion_;
public:
OutOfLineTruncateSlow(FloatRegister src, Register dest, bool needFloat32Conversion = false)
: src_(src), dest_(dest), needFloat32Conversion_(needFloat32Conversion)
{ }
void accept(CodeGeneratorShared* codegen) {
codegen->visitOutOfLineTruncateSlow(this);
}
FloatRegister src() const {
return src_;
}
Register dest() const {
return dest_;
}
bool needFloat32Conversion() const {
return needFloat32Conversion_;
}
};
OutOfLineCode*
CodeGeneratorShared::oolTruncateDouble(FloatRegister src, Register dest, MInstruction* mir)
{
OutOfLineTruncateSlow* ool = new(alloc()) OutOfLineTruncateSlow(src, dest);
addOutOfLineCode(ool, mir);
return ool;
}
void
CodeGeneratorShared::emitTruncateDouble(FloatRegister src, Register dest, MInstruction* mir)
{
OutOfLineCode* ool = oolTruncateDouble(src, dest, mir);
masm.branchTruncateDouble(src, dest, ool->entry());
masm.bind(ool->rejoin());
}
void
CodeGeneratorShared::emitTruncateFloat32(FloatRegister src, Register dest, MInstruction* mir)
{
OutOfLineTruncateSlow* ool = new(alloc()) OutOfLineTruncateSlow(src, dest, true);
addOutOfLineCode(ool, mir);
masm.branchTruncateFloat32(src, dest, ool->entry());
masm.bind(ool->rejoin());
}
void
CodeGeneratorShared::visitOutOfLineTruncateSlow(OutOfLineTruncateSlow* ool)
{
FloatRegister src = ool->src();
Register dest = ool->dest();
saveVolatile(dest);
#if defined(JS_CODEGEN_ARM) || defined(JS_CODEGEN_ARM64)
if (ool->needFloat32Conversion()) {
masm.convertFloat32ToDouble(src, ScratchDoubleReg);
src = ScratchDoubleReg;
}
#else
FloatRegister srcSingle = src.asSingle();
if (ool->needFloat32Conversion()) {
MOZ_ASSERT(src.isSingle());
masm.push(src);
masm.convertFloat32ToDouble(src, src);
src = src.asDouble();
}
#endif
masm.setupUnalignedABICall(dest);
masm.passABIArg(src, MoveOp::DOUBLE);
if (gen->compilingAsmJS())
masm.callWithABI(wasm::SymbolicAddress::ToInt32);
else
masm.callWithABI(BitwiseCast<void*, int32_t(*)(double)>(JS::ToInt32));
masm.storeCallResult(dest);
#if !defined(JS_CODEGEN_ARM) && !defined(JS_CODEGEN_ARM64)
if (ool->needFloat32Conversion())
masm.pop(srcSingle);
#endif
restoreVolatile(dest);
masm.jump(ool->rejoin());
}
bool
CodeGeneratorShared::omitOverRecursedCheck() const
{
// If the current function makes no calls (which means it isn't recursive)
// and it uses only a small amount of stack space, it doesn't need a
// stack overflow check. Note that the actual number here is somewhat
// arbitrary, and codegen actually uses small bounded amounts of
// additional stack space in some cases too.
return frameSize() < 64 && !gen->performsCall();
}
void
CodeGeneratorShared::emitAsmJSCall(LAsmJSCall* ins)
{
MAsmJSCall* mir = ins->mir();
if (mir->spIncrement())
masm.freeStack(mir->spIncrement());
MOZ_ASSERT((sizeof(AsmJSFrame) + masm.framePushed()) % AsmJSStackAlignment == 0);
#ifdef DEBUG
static_assert(AsmJSStackAlignment >= ABIStackAlignment &&
AsmJSStackAlignment % ABIStackAlignment == 0,
"The asm.js stack alignment should subsume the ABI-required alignment");
Label ok;
masm.branchTestStackPtr(Assembler::Zero, Imm32(AsmJSStackAlignment - 1), &ok);
masm.breakpoint();
masm.bind(&ok);
#endif
MAsmJSCall::Callee callee = mir->callee();
switch (callee.which()) {
case MAsmJSCall::Callee::Internal:
masm.call(mir->desc(), callee.internal());
break;
case MAsmJSCall::Callee::Dynamic:
masm.call(mir->desc(), ToRegister(ins->getOperand(mir->dynamicCalleeOperandIndex())));
break;
case MAsmJSCall::Callee::Builtin:
masm.call(BuiltinToImmediate(callee.builtin()));
break;
}
if (mir->spIncrement())
masm.reserveStack(mir->spIncrement());
}
void
CodeGeneratorShared::emitPreBarrier(Register base, const LAllocation* index)
{
if (index->isConstant()) {
Address address(base, ToInt32(index) * sizeof(Value));
masm.patchableCallPreBarrier(address, MIRType_Value);
} else {
BaseIndex address(base, ToRegister(index), TimesEight);
masm.patchableCallPreBarrier(address, MIRType_Value);
}
}
void
CodeGeneratorShared::emitPreBarrier(Address address)
{
masm.patchableCallPreBarrier(address, MIRType_Value);
}
Label*
CodeGeneratorShared::labelForBackedgeWithImplicitCheck(MBasicBlock* mir)
{
// If this is a loop backedge to a loop header with an implicit interrupt
// check, use a patchable jump. Skip this search if compiling without a
// script for asm.js, as there will be no interrupt check instruction.
// Due to critical edge unsplitting there may no longer be unique loop
// backedges, so just look for any edge going to an earlier block in RPO.
if (!gen->compilingAsmJS() && mir->isLoopHeader() && mir->id() <= current->mir()->id()) {
for (LInstructionIterator iter = mir->lir()->begin(); iter != mir->lir()->end(); iter++) {
if (iter->isMoveGroup()) {
// Continue searching for an interrupt check.
} else if (iter->isInterruptCheckImplicit()) {
return iter->toInterruptCheckImplicit()->oolEntry();
} else {
// The interrupt check should be the first instruction in the
// loop header other than the initial label and move groups.
MOZ_ASSERT(iter->isInterruptCheck());
return nullptr;
}
}
}
return nullptr;
}
void
CodeGeneratorShared::jumpToBlock(MBasicBlock* mir)
{
// Skip past trivial blocks.
mir = skipTrivialBlocks(mir);
// No jump necessary if we can fall through to the next block.
if (isNextBlock(mir->lir()))
return;
if (Label* oolEntry = labelForBackedgeWithImplicitCheck(mir)) {
// Note: the backedge is initially a jump to the next instruction.
// It will be patched to the target block's label during link().
RepatchLabel rejoin;
CodeOffsetJump backedge = masm.backedgeJump(&rejoin, mir->lir()->label());
masm.bind(&rejoin);
masm.propagateOOM(patchableBackedges_.append(PatchableBackedgeInfo(backedge, mir->lir()->label(), oolEntry)));
} else {
masm.jump(mir->lir()->label());
}
}
// This function is not used for MIPS/MIPS64. MIPS has branchToBlock.
#if !defined(JS_CODEGEN_MIPS32) && !defined(JS_CODEGEN_MIPS64)
void
CodeGeneratorShared::jumpToBlock(MBasicBlock* mir, Assembler::Condition cond)
{
// Skip past trivial blocks.
mir = skipTrivialBlocks(mir);
if (Label* oolEntry = labelForBackedgeWithImplicitCheck(mir)) {
// Note: the backedge is initially a jump to the next instruction.
// It will be patched to the target block's label during link().
RepatchLabel rejoin;
CodeOffsetJump backedge = masm.jumpWithPatch(&rejoin, cond, mir->lir()->label());
masm.bind(&rejoin);
masm.propagateOOM(patchableBackedges_.append(PatchableBackedgeInfo(backedge, mir->lir()->label(), oolEntry)));
} else {
masm.j(cond, mir->lir()->label());
}
}
#endif
MOZ_WARN_UNUSED_RESULT bool
CodeGeneratorShared::addCacheLocations(const CacheLocationList& locs, size_t* numLocs,
size_t* curIndex)
{
size_t firstIndex = runtimeData_.length();
size_t numLocations = 0;
for (CacheLocationList::iterator iter = locs.begin(); iter != locs.end(); iter++) {
// allocateData() ensures that sizeof(CacheLocation) is word-aligned.
// If this changes, we will need to pad to ensure alignment.
if (!allocateData(sizeof(CacheLocation), curIndex))
return false;
new (&runtimeData_[*curIndex]) CacheLocation(iter->pc, iter->script);
numLocations++;
}
MOZ_ASSERT(numLocations != 0);
*numLocs = numLocations;
*curIndex = firstIndex;
return true;
}
ReciprocalMulConstants
CodeGeneratorShared::computeDivisionConstants(uint32_t d, int maxLog) {
MOZ_ASSERT(maxLog >= 2 && maxLog <= 32);
// In what follows, 0 < d < 2^maxLog and d is not a power of 2.
MOZ_ASSERT(d < (uint64_t(1) << maxLog) && (d & (d - 1)) != 0);
// Speeding up division by non power-of-2 constants is possible by
// calculating, during compilation, a value M such that high-order
// bits of M*n correspond to the result of the division of n by d.
// No value of M can serve this purpose for arbitrarily big values
// of n but, for optimizing integer division, we're just concerned
// with values of n whose absolute value is bounded (by fitting in
// an integer type, say). With this in mind, we'll find a constant
// M as above that works for -2^maxLog <= n < 2^maxLog; maxLog can
// then be 31 for signed division or 32 for unsigned division.
//
// The original presentation of this technique appears in Hacker's
// Delight, a book by Henry S. Warren, Jr.. A proof of correctness
// for our version follows; we'll denote maxLog by L in the proof,
// for conciseness.
//
// Formally, for |d| < 2^L, we'll compute two magic values M and s
// in the ranges 0 <= M < 2^(L+1) and 0 <= s <= L such that
// (M * n) >> (32 + s) = floor(n/d) if 0 <= n < 2^L
// (M * n) >> (32 + s) = ceil(n/d) - 1 if -2^L <= n < 0.
//
// Define p = 32 + s, M = ceil(2^p/d), and assume that s satisfies
// M - 2^p/d <= 2^(p-L)/d. (1)
// (Observe that p = CeilLog32(d) + L satisfies this, as the right
// side of (1) is at least one in this case). Then,
//
// a) If p <= CeilLog32(d) + L, then M < 2^(L+1) - 1.
// Proof: Indeed, M is monotone in p and, for p equal to the above
// value, the bounds 2^L > d >= 2^(p-L-1) + 1 readily imply that
// 2^p / d < 2^p/(d - 1) * (d - 1)/d
// <= 2^(L+1) * (1 - 1/d) < 2^(L+1) - 2.
// The claim follows by applying the ceiling function.
//
// b) For any 0 <= n < 2^L, floor(Mn/2^p) = floor(n/d).
// Proof: Put x = floor(Mn/2^p); it's the unique integer for which
// Mn/2^p - 1 < x <= Mn/2^p. (2)
// Using M >= 2^p/d on the LHS and (1) on the RHS, we get
// n/d - 1 < x <= n/d + n/(2^L d) < n/d + 1/d.
// Since x is an integer, it's not in the interval (n/d, (n+1)/d),
// and so n/d - 1 < x <= n/d, which implies x = floor(n/d).
//
// c) For any -2^L <= n < 0, floor(Mn/2^p) + 1 = ceil(n/d).
// Proof: The proof is similar. Equation (2) holds as above. Using
// M > 2^p/d (d isn't a power of 2) on the RHS and (1) on the LHS,
// n/d + n/(2^L d) - 1 < x < n/d.
// Using n >= -2^L and summing 1,
// n/d - 1/d < x + 1 < n/d + 1.
// Since x + 1 is an integer, this implies n/d <= x + 1 < n/d + 1.
// In other words, x + 1 = ceil(n/d).
//
// Condition (1) isn't necessary for the existence of M and s with
// the properties above. Hacker's Delight provides a slightly less
// restrictive condition when d >= 196611, at the cost of a 3-page
// proof of correctness, for the case L = 31.
//
// Note that, since d*M - 2^p = d - (2^p)%d, (1) can be written as
// 2^(p-L) >= d - (2^p)%d.
// In order to avoid overflow in the (2^p) % d calculation, we can
// compute it as (2^p-1) % d + 1, where 2^p-1 can then be computed
// without overflow as UINT64_MAX >> (64-p).
// We now compute the least p >= 32 with the property above...
int32_t p = 32;
while ((uint64_t(1) << (p-maxLog)) + (UINT64_MAX >> (64-p)) % d + 1 < d)
p++;
// ...and the corresponding M. For either the signed (L=31) or the
// unsigned (L=32) case, this value can be too large (cf. item a).
// Codegen can still multiply by M by multiplying by (M - 2^L) and
// adjusting the value afterwards, if this is the case.
ReciprocalMulConstants rmc;
rmc.multiplier = (UINT64_MAX >> (64-p))/d + 1;
rmc.shiftAmount = p - 32;
return rmc;
}
#ifdef JS_TRACE_LOGGING
void
CodeGeneratorShared::emitTracelogScript(bool isStart)
{
if (!TraceLogTextIdEnabled(TraceLogger_Scripts))
return;
Label done;
AllocatableRegisterSet regs(RegisterSet::Volatile());
Register logger = regs.takeAnyGeneral();
Register script = regs.takeAnyGeneral();
masm.Push(logger);
CodeOffset patchLogger = masm.movWithPatch(ImmPtr(nullptr), logger);
masm.propagateOOM(patchableTraceLoggers_.append(patchLogger));
Address enabledAddress(logger, TraceLoggerThread::offsetOfEnabled());
masm.branch32(Assembler::Equal, enabledAddress, Imm32(0), &done);
masm.Push(script);
CodeOffset patchScript = masm.movWithPatch(ImmWord(0), script);
masm.propagateOOM(patchableTLScripts_.append(patchScript));
if (isStart)
masm.tracelogStartId(logger, script);
else
masm.tracelogStopId(logger, script);
masm.Pop(script);
masm.bind(&done);
masm.Pop(logger);
}
void
CodeGeneratorShared::emitTracelogTree(bool isStart, uint32_t textId)
{
if (!TraceLogTextIdEnabled(textId))
return;
Label done;
AllocatableRegisterSet regs(RegisterSet::Volatile());
Register logger = regs.takeAnyGeneral();
masm.Push(logger);
CodeOffset patchLocation = masm.movWithPatch(ImmPtr(nullptr), logger);
masm.propagateOOM(patchableTraceLoggers_.append(patchLocation));
Address enabledAddress(logger, TraceLoggerThread::offsetOfEnabled());
masm.branch32(Assembler::Equal, enabledAddress, Imm32(0), &done);
if (isStart)
masm.tracelogStartId(logger, textId);
else
masm.tracelogStopId(logger, textId);
masm.bind(&done);
masm.Pop(logger);
}
#endif
} // namespace jit
} // namespace js