Mypal/js/src/jit/EffectiveAddressAnalysis.cpp
2021-02-04 16:48:36 +02:00

277 lines
8.9 KiB
C++

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
* 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/EffectiveAddressAnalysis.h"
#include "jit/IonAnalysis.h"
#include "jit/MIR.h"
#include "jit/MIRGraph.h"
using namespace js;
using namespace jit;
static void
AnalyzeLsh(TempAllocator& alloc, MLsh* lsh)
{
if (lsh->specialization() != MIRType::Int32)
return;
if (lsh->isRecoveredOnBailout())
return;
MDefinition* index = lsh->lhs();
MOZ_ASSERT(index->type() == MIRType::Int32);
MConstant* shiftValue = lsh->rhs()->maybeConstantValue();
if (!shiftValue)
return;
if (shiftValue->type() != MIRType::Int32 || !IsShiftInScaleRange(shiftValue->toInt32()))
return;
Scale scale = ShiftToScale(shiftValue->toInt32());
int32_t displacement = 0;
MInstruction* last = lsh;
MDefinition* base = nullptr;
while (true) {
if (!last->hasOneUse())
break;
MUseIterator use = last->usesBegin();
if (!use->consumer()->isDefinition() || !use->consumer()->toDefinition()->isAdd())
break;
MAdd* add = use->consumer()->toDefinition()->toAdd();
if (add->specialization() != MIRType::Int32 || !add->isTruncated())
break;
MDefinition* other = add->getOperand(1 - add->indexOf(*use));
if (MConstant* otherConst = other->maybeConstantValue()) {
displacement += otherConst->toInt32();
} else {
if (base)
break;
base = other;
}
last = add;
if (last->isRecoveredOnBailout())
return;
}
if (!base) {
uint32_t elemSize = 1 << ScaleToShift(scale);
if (displacement % elemSize != 0)
return;
if (!last->hasOneUse())
return;
MUseIterator use = last->usesBegin();
if (!use->consumer()->isDefinition() || !use->consumer()->toDefinition()->isBitAnd())
return;
MBitAnd* bitAnd = use->consumer()->toDefinition()->toBitAnd();
if (bitAnd->isRecoveredOnBailout())
return;
MDefinition* other = bitAnd->getOperand(1 - bitAnd->indexOf(*use));
MConstant* otherConst = other->maybeConstantValue();
if (!otherConst || otherConst->type() != MIRType::Int32)
return;
uint32_t bitsClearedByShift = elemSize - 1;
uint32_t bitsClearedByMask = ~uint32_t(otherConst->toInt32());
if ((bitsClearedByShift & bitsClearedByMask) != bitsClearedByMask)
return;
bitAnd->replaceAllUsesWith(last);
return;
}
if (base->isRecoveredOnBailout())
return;
MEffectiveAddress* eaddr = MEffectiveAddress::New(alloc, base, index, scale, displacement);
last->replaceAllUsesWith(eaddr);
last->block()->insertAfter(last, eaddr);
}
// Transform:
//
// [AddI]
// addl $9, %esi
// [LoadUnboxedScalar]
// movsd 0x0(%rbx,%rsi,8), %xmm4
//
// into:
//
// [LoadUnboxedScalar]
// movsd 0x48(%rbx,%rsi,8), %xmm4
//
// This is possible when the AddI is only used by the LoadUnboxedScalar opcode.
static void
AnalyzeLoadUnboxedScalar(TempAllocator& alloc, MLoadUnboxedScalar* load)
{
if (load->isRecoveredOnBailout())
return;
if (!load->getOperand(1)->isAdd())
return;
JitSpew(JitSpew_EAA, "analyze: %s%u", load->opName(), load->id());
MAdd* add = load->getOperand(1)->toAdd();
if (add->specialization() != MIRType::Int32 || !add->hasUses() ||
add->truncateKind() != MDefinition::TruncateKind::Truncate)
{
return;
}
MDefinition* lhs = add->lhs();
MDefinition* rhs = add->rhs();
MDefinition* constant = nullptr;
MDefinition* node = nullptr;
if (lhs->isConstant()) {
constant = lhs;
node = rhs;
} else if (rhs->isConstant()) {
constant = rhs;
node = lhs;
} else
return;
MOZ_ASSERT(constant->type() == MIRType::Int32);
size_t storageSize = Scalar::byteSize(load->storageType());
int32_t c1 = load->offsetAdjustment();
int32_t c2 = 0;
if (!SafeMul(constant->maybeConstantValue()->toInt32(), storageSize, &c2))
return;
int32_t offset = 0;
if (!SafeAdd(c1, c2, &offset))
return;
JitSpew(JitSpew_EAA, "set offset: %d + %d = %d on: %s%u", c1, c2, offset,
load->opName(), load->id());
load->setOffsetAdjustment(offset);
load->replaceOperand(1, node);
if (!add->hasLiveDefUses() && DeadIfUnused(add) && add->canRecoverOnBailout()) {
JitSpew(JitSpew_EAA, "mark as recovered on bailout: %s%u",
add->opName(), add->id());
add->setRecoveredOnBailoutUnchecked();
}
}
template<typename AsmJSMemoryAccess>
bool
EffectiveAddressAnalysis::tryAddDisplacement(AsmJSMemoryAccess* ins, int32_t o)
{
#ifdef WASM_HUGE_MEMORY
// Compute the new offset. Check for overflow.
uint32_t oldOffset = ins->offset();
uint32_t newOffset = oldOffset + o;
if (o < 0 ? (newOffset >= oldOffset) : (newOffset < oldOffset))
return false;
// The offset must ultimately be written into the offset immediate of a load
// or store instruction so don't allow folding of the offset is bigger.
if (newOffset >= wasm::OffsetGuardLimit)
return false;
// Everything checks out. This is the new offset.
ins->setOffset(newOffset);
return true;
#else
return false;
#endif
}
template<typename AsmJSMemoryAccess>
void
EffectiveAddressAnalysis::analyzeAsmJSHeapAccess(AsmJSMemoryAccess* ins)
{
MDefinition* base = ins->base();
if (base->isConstant()) {
// Look for heap[i] where i is a constant offset, and fold the offset.
// By doing the folding now, we simplify the task of codegen; the offset
// is always the address mode immediate. This also allows it to avoid
// a situation where the sum of a constant pointer value and a non-zero
// offset doesn't actually fit into the address mode immediate.
int32_t imm = base->toConstant()->toInt32();
if (imm != 0 && tryAddDisplacement(ins, imm)) {
MInstruction* zero = MConstant::New(graph_.alloc(), Int32Value(0));
ins->block()->insertBefore(ins, zero);
ins->replaceBase(zero);
}
// If the index is within the minimum heap length, we can optimize
// away the bounds check.
if (imm >= 0) {
int32_t end = (uint32_t)imm + ins->byteSize();
if (end >= imm && (uint32_t)end <= mir_->minWasmHeapLength())
ins->removeBoundsCheck();
}
} else if (base->isAdd()) {
// Look for heap[a+i] where i is a constant offset, and fold the offset.
// Alignment masks have already been moved out of the way by the
// Alignment Mask Analysis pass.
MDefinition* op0 = base->toAdd()->getOperand(0);
MDefinition* op1 = base->toAdd()->getOperand(1);
if (op0->isConstant())
mozilla::Swap(op0, op1);
if (op1->isConstant()) {
int32_t imm = op1->toConstant()->toInt32();
if (tryAddDisplacement(ins, imm))
ins->replaceBase(op0);
}
}
}
// This analysis converts patterns of the form:
// truncate(x + (y << {0,1,2,3}))
// truncate(x + (y << {0,1,2,3}) + imm32)
// into a single lea instruction, and patterns of the form:
// asmload(x + imm32)
// asmload(x << {0,1,2,3})
// asmload((x << {0,1,2,3}) + imm32)
// asmload((x << {0,1,2,3}) & mask) (where mask is redundant with shift)
// asmload(((x << {0,1,2,3}) + imm32) & mask) (where mask is redundant with shift + imm32)
// into a single asmload instruction (and for asmstore too).
//
// Additionally, we should consider the general forms:
// truncate(x + y + imm32)
// truncate((y << {0,1,2,3}) + imm32)
bool
EffectiveAddressAnalysis::analyze()
{
for (ReversePostorderIterator block(graph_.rpoBegin()); block != graph_.rpoEnd(); block++) {
for (MInstructionIterator i = block->begin(); i != block->end(); i++) {
if (!graph_.alloc().ensureBallast())
return false;
// Note that we don't check for MAsmJSCompareExchangeHeap
// or MAsmJSAtomicBinopHeap, because the backend and the OOB
// mechanism don't support non-zero offsets for them yet
// (TODO bug 1254935).
if (i->isLsh())
AnalyzeLsh(graph_.alloc(), i->toLsh());
else if (i->isLoadUnboxedScalar())
AnalyzeLoadUnboxedScalar(graph_.alloc(), i->toLoadUnboxedScalar());
else if (i->isAsmJSLoadHeap())
analyzeAsmJSHeapAccess(i->toAsmJSLoadHeap());
else if (i->isAsmJSStoreHeap())
analyzeAsmJSHeapAccess(i->toAsmJSStoreHeap());
}
}
return true;
}