blob: 38ed98adafd1124c44582a2524ab362992a2f178 [file] [log] [blame]
/*
* Copyright (C) 2017 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "load_store_analysis.h"
#include "base/scoped_arena_allocator.h"
#include "optimizing/escape.h"
namespace art {
// A cap for the number of heap locations to prevent pathological time/space consumption.
// The number of heap locations for most of the methods stays below this threshold.
constexpr size_t kMaxNumberOfHeapLocations = 32;
// Test if two integer ranges [l1,h1] and [l2,h2] overlap.
// Note that the ranges are inclusive on both ends.
// l1|------|h1
// l2|------|h2
static bool CanIntegerRangesOverlap(int64_t l1, int64_t h1, int64_t l2, int64_t h2) {
return std::max(l1, l2) <= std::min(h1, h2);
}
static bool CanBinaryOpAndIndexAlias(const HBinaryOperation* idx1,
const size_t vector_length1,
const HInstruction* idx2,
const size_t vector_length2) {
if (!IsAddOrSub(idx1)) {
// We currently only support Add and Sub operations.
return true;
}
if (idx1->AsBinaryOperation()->GetLeastConstantLeft() != idx2) {
// Cannot analyze [i+CONST1] and [j].
return true;
}
if (!idx1->GetConstantRight()->IsIntConstant()) {
return true;
}
// Since 'i' are the same in [i+CONST] and [i],
// further compare [CONST] and [0].
int64_t l1 = idx1->IsAdd() ?
idx1->GetConstantRight()->AsIntConstant()->GetValue() :
-idx1->GetConstantRight()->AsIntConstant()->GetValue();
int64_t l2 = 0;
int64_t h1 = l1 + (vector_length1 - 1);
int64_t h2 = l2 + (vector_length2 - 1);
return CanIntegerRangesOverlap(l1, h1, l2, h2);
}
static bool CanBinaryOpsAlias(const HBinaryOperation* idx1,
const size_t vector_length1,
const HBinaryOperation* idx2,
const size_t vector_length2) {
if (!IsAddOrSub(idx1) || !IsAddOrSub(idx2)) {
// We currently only support Add and Sub operations.
return true;
}
if (idx1->AsBinaryOperation()->GetLeastConstantLeft() !=
idx2->AsBinaryOperation()->GetLeastConstantLeft()) {
// Cannot analyze [i+CONST1] and [j+CONST2].
return true;
}
if (!idx1->GetConstantRight()->IsIntConstant() ||
!idx2->GetConstantRight()->IsIntConstant()) {
return true;
}
// Since 'i' are the same in [i+CONST1] and [i+CONST2],
// further compare [CONST1] and [CONST2].
int64_t l1 = idx1->IsAdd() ?
idx1->GetConstantRight()->AsIntConstant()->GetValue() :
-idx1->GetConstantRight()->AsIntConstant()->GetValue();
int64_t l2 = idx2->IsAdd() ?
idx2->GetConstantRight()->AsIntConstant()->GetValue() :
-idx2->GetConstantRight()->AsIntConstant()->GetValue();
int64_t h1 = l1 + (vector_length1 - 1);
int64_t h2 = l2 + (vector_length2 - 1);
return CanIntegerRangesOverlap(l1, h1, l2, h2);
}
// Make sure we mark any writes/potential writes to heap-locations within partially
// escaped values as escaping.
void ReferenceInfo::PrunePartialEscapeWrites() {
if (!subgraph_.IsValid()) {
// All paths escape.
return;
}
HGraph* graph = reference_->GetBlock()->GetGraph();
ArenaBitVector additional_exclusions(
allocator_, graph->GetBlocks().size(), false, kArenaAllocLSA);
for (const HUseListNode<HInstruction*>& use : reference_->GetUses()) {
const HInstruction* user = use.GetUser();
if (!additional_exclusions.IsBitSet(user->GetBlock()->GetBlockId()) &&
subgraph_.ContainsBlock(user->GetBlock()) &&
(user->IsUnresolvedInstanceFieldSet() || user->IsUnresolvedStaticFieldSet() ||
user->IsInstanceFieldSet() || user->IsStaticFieldSet() || user->IsArraySet()) &&
(reference_ == user->InputAt(0)) &&
std::any_of(subgraph_.UnreachableBlocks().begin(),
subgraph_.UnreachableBlocks().end(),
[&](const HBasicBlock* excluded) -> bool {
return reference_->GetBlock()->GetGraph()->PathBetween(excluded,
user->GetBlock());
})) {
// This object had memory written to it somewhere, if it escaped along
// some paths prior to the current block this write also counts as an
// escape.
additional_exclusions.SetBit(user->GetBlock()->GetBlockId());
}
}
if (UNLIKELY(additional_exclusions.IsAnyBitSet())) {
for (uint32_t exc : additional_exclusions.Indexes()) {
subgraph_.RemoveBlock(graph->GetBlocks()[exc]);
}
}
}
bool HeapLocationCollector::InstructionEligibleForLSERemoval(HInstruction* inst) const {
if (inst->IsNewInstance()) {
return !inst->AsNewInstance()->NeedsChecks();
} else if (inst->IsNewArray()) {
HInstruction* array_length = inst->AsNewArray()->GetLength();
bool known_array_length =
array_length->IsIntConstant() && array_length->AsIntConstant()->GetValue() >= 0;
return known_array_length &&
std::all_of(inst->GetUses().cbegin(),
inst->GetUses().cend(),
[&](const HUseListNode<HInstruction*>& user) {
if (user.GetUser()->IsArrayGet() || user.GetUser()->IsArraySet()) {
return user.GetUser()->InputAt(1)->IsIntConstant();
}
return true;
});
} else {
return false;
}
}
void ReferenceInfo::CollectPartialEscapes(HGraph* graph) {
ScopedArenaAllocator saa(graph->GetArenaStack());
ArenaBitVector seen_instructions(&saa, graph->GetCurrentInstructionId(), false, kArenaAllocLSA);
// Get regular escapes.
ScopedArenaVector<HInstruction*> additional_escape_vectors(saa.Adapter(kArenaAllocLSA));
LambdaEscapeVisitor scan_instructions([&](HInstruction* escape) -> bool {
HandleEscape(escape);
// LSE can't track heap-locations through Phi and Select instructions so we
// need to assume all escapes from these are escapes for the base reference.
if ((escape->IsPhi() || escape->IsSelect()) && !seen_instructions.IsBitSet(escape->GetId())) {
seen_instructions.SetBit(escape->GetId());
additional_escape_vectors.push_back(escape);
}
return true;
});
additional_escape_vectors.push_back(reference_);
while (!additional_escape_vectors.empty()) {
HInstruction* ref = additional_escape_vectors.back();
additional_escape_vectors.pop_back();
DCHECK(ref == reference_ || ref->IsPhi() || ref->IsSelect()) << *ref;
VisitEscapes(ref, scan_instructions);
}
// Mark irreducible loop headers as escaping since they cannot be tracked through.
for (HBasicBlock* blk : graph->GetActiveBlocks()) {
if (blk->IsLoopHeader() && blk->GetLoopInformation()->IsIrreducible()) {
HandleEscape(blk);
}
}
}
void HeapLocationCollector::DumpReferenceStats(OptimizingCompilerStats* stats) {
if (stats == nullptr) {
return;
}
std::vector<bool> seen_instructions(GetGraph()->GetCurrentInstructionId(), false);
for (auto hl : heap_locations_) {
auto ri = hl->GetReferenceInfo();
if (ri == nullptr || seen_instructions[ri->GetReference()->GetId()]) {
continue;
}
auto instruction = ri->GetReference();
seen_instructions[instruction->GetId()] = true;
if (ri->IsSingletonAndRemovable()) {
if (InstructionEligibleForLSERemoval(instruction)) {
MaybeRecordStat(stats, MethodCompilationStat::kFullLSEPossible);
}
}
// TODO This is an estimate of the number of allocations we will be able
// to (partially) remove. As additional work is done this can be refined.
if (ri->IsPartialSingleton() && instruction->IsNewInstance() &&
ri->GetNoEscapeSubgraph()->ContainsBlock(instruction->GetBlock()) &&
!ri->GetNoEscapeSubgraph()->GetExcludedCohorts().empty() &&
InstructionEligibleForLSERemoval(instruction)) {
MaybeRecordStat(stats, MethodCompilationStat::kPartialLSEPossible);
}
}
}
bool HeapLocationCollector::CanArrayElementsAlias(const HInstruction* idx1,
const size_t vector_length1,
const HInstruction* idx2,
const size_t vector_length2) const {
DCHECK(idx1 != nullptr);
DCHECK(idx2 != nullptr);
DCHECK_GE(vector_length1, HeapLocation::kScalar);
DCHECK_GE(vector_length2, HeapLocation::kScalar);
// [i] and [i].
if (idx1 == idx2) {
return true;
}
// [CONST1] and [CONST2].
if (idx1->IsIntConstant() && idx2->IsIntConstant()) {
int64_t l1 = idx1->AsIntConstant()->GetValue();
int64_t l2 = idx2->AsIntConstant()->GetValue();
// To avoid any overflow in following CONST+vector_length calculation,
// use int64_t instead of int32_t.
int64_t h1 = l1 + (vector_length1 - 1);
int64_t h2 = l2 + (vector_length2 - 1);
return CanIntegerRangesOverlap(l1, h1, l2, h2);
}
// [i+CONST] and [i].
if (idx1->IsBinaryOperation() &&
idx1->AsBinaryOperation()->GetConstantRight() != nullptr &&
idx1->AsBinaryOperation()->GetLeastConstantLeft() == idx2) {
return CanBinaryOpAndIndexAlias(idx1->AsBinaryOperation(),
vector_length1,
idx2,
vector_length2);
}
// [i] and [i+CONST].
if (idx2->IsBinaryOperation() &&
idx2->AsBinaryOperation()->GetConstantRight() != nullptr &&
idx2->AsBinaryOperation()->GetLeastConstantLeft() == idx1) {
return CanBinaryOpAndIndexAlias(idx2->AsBinaryOperation(),
vector_length2,
idx1,
vector_length1);
}
// [i+CONST1] and [i+CONST2].
if (idx1->IsBinaryOperation() &&
idx1->AsBinaryOperation()->GetConstantRight() != nullptr &&
idx2->IsBinaryOperation() &&
idx2->AsBinaryOperation()->GetConstantRight() != nullptr) {
return CanBinaryOpsAlias(idx1->AsBinaryOperation(),
vector_length1,
idx2->AsBinaryOperation(),
vector_length2);
}
// By default, MAY alias.
return true;
}
bool LoadStoreAnalysis::Run() {
for (HBasicBlock* block : graph_->GetReversePostOrder()) {
heap_location_collector_.VisitBasicBlock(block);
}
if (heap_location_collector_.GetNumberOfHeapLocations() > kMaxNumberOfHeapLocations) {
// Bail out if there are too many heap locations to deal with.
heap_location_collector_.CleanUp();
return false;
}
if (!heap_location_collector_.HasHeapStores()) {
// Without heap stores, this pass would act mostly as GVN on heap accesses.
heap_location_collector_.CleanUp();
return false;
}
if (heap_location_collector_.HasVolatile() || heap_location_collector_.HasMonitorOps()) {
// Don't do load/store elimination if the method has volatile field accesses or
// monitor operations, for now.
// TODO: do it right.
heap_location_collector_.CleanUp();
return false;
}
heap_location_collector_.BuildAliasingMatrix();
heap_location_collector_.DumpReferenceStats(stats_);
return true;
}
} // namespace art