Determine HLoadClass/String load kind early.
This helps save memory by avoiding the allocation of
HEnvironment and related objects for AOT references to
boot image strings and classes (kBootImage* load kinds)
and also for JIT references (kJitTableAddress).
Compiling aosp_taimen-userdebug boot image, the most memory
hungry method BatteryStats.dumpLocked() needs
- before:
Used 55105384 bytes of arena memory...
...
UseListNode 10009704
Environment 423248
EnvVRegs 20676560
...
- after:
Used 50559176 bytes of arena memory...
...
UseListNode 8568936
Environment 365680
EnvVRegs 17628704
...
Test: m test-art-host-gtest
Test: testrunner.py --host --optimizing --jit
Bug: 34053922
Change-Id: I68e73a438e6ac8e8908e6fccf53bbeea8a64a077
diff --git a/compiler/optimizing/inliner.cc b/compiler/optimizing/inliner.cc
index 560372e..a175c21 100644
--- a/compiler/optimizing/inliner.cc
+++ b/compiler/optimizing/inliner.cc
@@ -876,9 +876,9 @@
load_class, codegen_, compiler_driver_, caller_compilation_unit_);
DCHECK(kind != HLoadClass::LoadKind::kInvalid)
<< "We should always be able to reference a class for inline caches";
- // Insert before setting the kind, as setting the kind affects the inputs.
- bb_cursor->InsertInstructionAfter(load_class, receiver_class);
+ // Load kind must be set before inserting the instruction into the graph.
load_class->SetLoadKind(kind);
+ bb_cursor->InsertInstructionAfter(load_class, receiver_class);
// In AOT mode, we will most likely load the class from BSS, which will involve a call
// to the runtime. In this case, the load instruction will need an environment so copy
// it from the invoke instruction.
@@ -1932,7 +1932,7 @@
// optimization that could lead to a HDeoptimize. The following optimizations do not.
HDeadCodeElimination dce(callee_graph, inline_stats_, "dead_code_elimination$inliner");
HConstantFolding fold(callee_graph, "constant_folding$inliner");
- HSharpening sharpening(callee_graph, codegen_, dex_compilation_unit, compiler_driver_, handles_);
+ HSharpening sharpening(callee_graph, codegen_, compiler_driver_);
InstructionSimplifier simplify(callee_graph, codegen_, compiler_driver_, inline_stats_);
IntrinsicsRecognizer intrinsics(callee_graph, inline_stats_);
diff --git a/compiler/optimizing/instruction_builder.cc b/compiler/optimizing/instruction_builder.cc
index 782546c..4485f06 100644
--- a/compiler/optimizing/instruction_builder.cc
+++ b/compiler/optimizing/instruction_builder.cc
@@ -1128,7 +1128,7 @@
MethodCompilationStat::kConstructorFenceGeneratedNew);
}
-static bool IsSubClass(mirror::Class* to_test, mirror::Class* super_class)
+static bool IsSubClass(ObjPtr<mirror::Class> to_test, ObjPtr<mirror::Class> super_class)
REQUIRES_SHARED(Locks::mutator_lock_) {
return to_test != nullptr && !to_test->IsInterface() && to_test->IsSubClass(super_class);
}
@@ -1424,7 +1424,7 @@
return true;
}
-static mirror::Class* GetClassFrom(CompilerDriver* driver,
+static ObjPtr<mirror::Class> GetClassFrom(CompilerDriver* driver,
const DexCompilationUnit& compilation_unit) {
ScopedObjectAccess soa(Thread::Current());
Handle<mirror::ClassLoader> class_loader = compilation_unit.GetClassLoader();
@@ -1433,11 +1433,11 @@
return driver->ResolveCompilingMethodsClass(soa, dex_cache, class_loader, &compilation_unit);
}
-mirror::Class* HInstructionBuilder::GetOutermostCompilingClass() const {
+ObjPtr<mirror::Class> HInstructionBuilder::GetOutermostCompilingClass() const {
return GetClassFrom(compiler_driver_, *outer_compilation_unit_);
}
-mirror::Class* HInstructionBuilder::GetCompilingClass() const {
+ObjPtr<mirror::Class> HInstructionBuilder::GetCompilingClass() const {
return GetClassFrom(compiler_driver_, *dex_compilation_unit_);
}
@@ -1799,6 +1799,17 @@
}
}
+void HInstructionBuilder::BuildLoadString(dex::StringIndex string_index, uint32_t dex_pc) {
+ HLoadString* load_string =
+ new (allocator_) HLoadString(graph_->GetCurrentMethod(), string_index, *dex_file_, dex_pc);
+ HSharpening::ProcessLoadString(load_string,
+ code_generator_,
+ compiler_driver_,
+ *dex_compilation_unit_,
+ handles_);
+ AppendInstruction(load_string);
+}
+
HLoadClass* HInstructionBuilder::BuildLoadClass(dex::TypeIndex type_index, uint32_t dex_pc) {
ScopedObjectAccess soa(Thread::Current());
const DexFile& dex_file = *dex_compilation_unit_->GetDexFile();
@@ -1811,7 +1822,7 @@
if (klass->IsPublic()) {
needs_access_check = false;
} else {
- mirror::Class* compiling_class = GetCompilingClass();
+ ObjPtr<mirror::Class> compiling_class = GetCompilingClass();
if (compiling_class != nullptr && compiling_class->CanAccess(klass.Get())) {
needs_access_check = false;
}
@@ -1856,9 +1867,9 @@
// We actually cannot reference this class, we're forced to bail.
return nullptr;
}
- // Append the instruction first, as setting the load kind affects the inputs.
- AppendInstruction(load_class);
+ // Load kind must be set before inserting the instruction into the graph.
load_class->SetLoadKind(load_kind);
+ AppendInstruction(load_class);
return load_class;
}
@@ -2837,20 +2848,14 @@
case Instruction::CONST_STRING: {
dex::StringIndex string_index(instruction.VRegB_21c());
- AppendInstruction(new (allocator_) HLoadString(graph_->GetCurrentMethod(),
- string_index,
- *dex_file_,
- dex_pc));
+ BuildLoadString(string_index, dex_pc);
UpdateLocal(instruction.VRegA_21c(), current_block_->GetLastInstruction());
break;
}
case Instruction::CONST_STRING_JUMBO: {
dex::StringIndex string_index(instruction.VRegB_31c());
- AppendInstruction(new (allocator_) HLoadString(graph_->GetCurrentMethod(),
- string_index,
- *dex_file_,
- dex_pc));
+ BuildLoadString(string_index, dex_pc);
UpdateLocal(instruction.VRegA_31c(), current_block_->GetLastInstruction());
break;
}
diff --git a/compiler/optimizing/instruction_builder.h b/compiler/optimizing/instruction_builder.h
index 2446ddb..0500d40 100644
--- a/compiler/optimizing/instruction_builder.h
+++ b/compiler/optimizing/instruction_builder.h
@@ -240,9 +240,10 @@
// Builds an instruction sequence for a switch statement.
void BuildSwitch(const Instruction& instruction, uint32_t dex_pc);
- // Builds a `HLoadClass` loading the given `type_index`. If `outer` is true,
- // this method will use the outer class's dex file to lookup the type at
- // `type_index`.
+ // Builds a `HLoadString` loading the given `string_index`.
+ void BuildLoadString(dex::StringIndex string_index, uint32_t dex_pc);
+
+ // Builds a `HLoadClass` loading the given `type_index`.
HLoadClass* BuildLoadClass(dex::TypeIndex type_index, uint32_t dex_pc);
HLoadClass* BuildLoadClass(dex::TypeIndex type_index,
@@ -253,10 +254,10 @@
REQUIRES_SHARED(Locks::mutator_lock_);
// Returns the outer-most compiling method's class.
- mirror::Class* GetOutermostCompilingClass() const;
+ ObjPtr<mirror::Class> GetOutermostCompilingClass() const;
// Returns the class whose method is being compiled.
- mirror::Class* GetCompilingClass() const;
+ ObjPtr<mirror::Class> GetCompilingClass() const;
// Returns whether `type_index` points to the outer-most compiling method's class.
bool IsOutermostCompilingClass(dex::TypeIndex type_index) const;
diff --git a/compiler/optimizing/nodes.cc b/compiler/optimizing/nodes.cc
index d117bfb..5f33ed6 100644
--- a/compiler/optimizing/nodes.cc
+++ b/compiler/optimizing/nodes.cc
@@ -2831,21 +2831,6 @@
}
}
-void HLoadClass::SetLoadKind(LoadKind load_kind) {
- SetPackedField<LoadKindField>(load_kind);
-
- if (load_kind != LoadKind::kRuntimeCall &&
- load_kind != LoadKind::kReferrersClass) {
- RemoveAsUserOfInput(0u);
- SetRawInputAt(0u, nullptr);
- }
-
- if (!NeedsEnvironment()) {
- RemoveEnvironment();
- SetSideEffects(SideEffects::None());
- }
-}
-
std::ostream& operator<<(std::ostream& os, HLoadClass::LoadKind rhs) {
switch (rhs) {
case HLoadClass::LoadKind::kReferrersClass:
@@ -2888,21 +2873,6 @@
}
}
-void HLoadString::SetLoadKind(LoadKind load_kind) {
- // Once sharpened, the load kind should not be changed again.
- DCHECK_EQ(GetLoadKind(), LoadKind::kRuntimeCall);
- SetPackedField<LoadKindField>(load_kind);
-
- if (load_kind != LoadKind::kRuntimeCall) {
- RemoveAsUserOfInput(0u);
- SetRawInputAt(0u, nullptr);
- }
- if (!NeedsEnvironment()) {
- RemoveEnvironment();
- SetSideEffects(SideEffects::None());
- }
-}
-
std::ostream& operator<<(std::ostream& os, HLoadString::LoadKind rhs) {
switch (rhs) {
case HLoadString::LoadKind::kBootImageLinkTimePcRelative:
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index 7fbd7f4..42a9d95 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -6061,6 +6061,20 @@
std::ostream& operator<<(std::ostream& os, HLoadClass::LoadKind rhs);
// Note: defined outside class to see operator<<(., HLoadClass::LoadKind).
+inline void HLoadClass::SetLoadKind(LoadKind load_kind) {
+ // The load kind should be determined before inserting the instruction to the graph.
+ DCHECK(GetBlock() == nullptr);
+ DCHECK(GetEnvironment() == nullptr);
+ SetPackedField<LoadKindField>(load_kind);
+ if (load_kind != LoadKind::kRuntimeCall && load_kind != LoadKind::kReferrersClass) {
+ special_input_ = HUserRecord<HInstruction*>(nullptr);
+ }
+ if (!NeedsEnvironment()) {
+ SetSideEffects(SideEffects::None());
+ }
+}
+
+// Note: defined outside class to see operator<<(., HLoadClass::LoadKind).
inline void HLoadClass::AddSpecialInput(HInstruction* special_input) {
// The special input is used for PC-relative loads on some architectures,
// including literal pool loads, which are PC-relative too.
@@ -6208,6 +6222,21 @@
std::ostream& operator<<(std::ostream& os, HLoadString::LoadKind rhs);
// Note: defined outside class to see operator<<(., HLoadString::LoadKind).
+inline void HLoadString::SetLoadKind(LoadKind load_kind) {
+ // The load kind should be determined before inserting the instruction to the graph.
+ DCHECK(GetBlock() == nullptr);
+ DCHECK(GetEnvironment() == nullptr);
+ DCHECK_EQ(GetLoadKind(), LoadKind::kRuntimeCall);
+ SetPackedField<LoadKindField>(load_kind);
+ if (load_kind != LoadKind::kRuntimeCall) {
+ special_input_ = HUserRecord<HInstruction*>(nullptr);
+ }
+ if (!NeedsEnvironment()) {
+ SetSideEffects(SideEffects::None());
+ }
+}
+
+// Note: defined outside class to see operator<<(., HLoadString::LoadKind).
inline void HLoadString::AddSpecialInput(HInstruction* special_input) {
// The special input is used for PC-relative loads on some architectures,
// including literal pool loads, which are PC-relative too.
diff --git a/compiler/optimizing/optimization.cc b/compiler/optimizing/optimization.cc
index 7edb642..7149d93 100644
--- a/compiler/optimizing/optimization.cc
+++ b/compiler/optimizing/optimization.cc
@@ -258,8 +258,7 @@
break;
}
case OptimizationPass::kSharpening:
- opt = new (allocator) HSharpening(
- graph, codegen, dex_compilation_unit, driver, handles, name);
+ opt = new (allocator) HSharpening(graph, codegen, driver, name);
break;
case OptimizationPass::kSelectGenerator:
opt = new (allocator) HSelectGenerator(graph, handles, stats, name);
diff --git a/compiler/optimizing/sharpening.cc b/compiler/optimizing/sharpening.cc
index e46c9a7..64092d3 100644
--- a/compiler/optimizing/sharpening.cc
+++ b/compiler/optimizing/sharpening.cc
@@ -45,8 +45,6 @@
SharpenInvokeStaticOrDirect(instruction->AsInvokeStaticOrDirect(),
codegen_,
compiler_driver_);
- } else if (instruction->IsLoadString()) {
- ProcessLoadString(instruction->AsLoadString());
}
// TODO: Move the sharpening of invoke-virtual/-interface/-super from HGraphBuilder
// here. Rewrite it to avoid the CompilerDriver's reliance on verifier data
@@ -147,10 +145,11 @@
invoke->SetDispatchInfo(dispatch_info);
}
-HLoadClass::LoadKind HSharpening::ComputeLoadClassKind(HLoadClass* load_class,
- CodeGenerator* codegen,
- CompilerDriver* compiler_driver,
- const DexCompilationUnit& dex_compilation_unit) {
+HLoadClass::LoadKind HSharpening::ComputeLoadClassKind(
+ HLoadClass* load_class,
+ CodeGenerator* codegen,
+ CompilerDriver* compiler_driver,
+ const DexCompilationUnit& dex_compilation_unit) {
Handle<mirror::Class> klass = load_class->GetClass();
DCHECK(load_class->GetLoadKind() == HLoadClass::LoadKind::kRuntimeCall ||
load_class->GetLoadKind() == HLoadClass::LoadKind::kReferrersClass)
@@ -237,7 +236,12 @@
return load_kind;
}
-void HSharpening::ProcessLoadString(HLoadString* load_string) {
+void HSharpening::ProcessLoadString(
+ HLoadString* load_string,
+ CodeGenerator* codegen,
+ CompilerDriver* compiler_driver,
+ const DexCompilationUnit& dex_compilation_unit,
+ VariableSizedHandleScope* handles) {
DCHECK_EQ(load_string->GetLoadKind(), HLoadString::LoadKind::kRuntimeCall);
const DexFile& dex_file = load_string->GetDexFile();
@@ -249,26 +253,26 @@
ClassLinker* class_linker = runtime->GetClassLinker();
ScopedObjectAccess soa(Thread::Current());
StackHandleScope<1> hs(soa.Self());
- Handle<mirror::DexCache> dex_cache = IsSameDexFile(dex_file, *compilation_unit_.GetDexFile())
- ? compilation_unit_.GetDexCache()
+ Handle<mirror::DexCache> dex_cache = IsSameDexFile(dex_file, *dex_compilation_unit.GetDexFile())
+ ? dex_compilation_unit.GetDexCache()
: hs.NewHandle(class_linker->FindDexCache(soa.Self(), dex_file));
- mirror::String* string = nullptr;
+ ObjPtr<mirror::String> string = nullptr;
- if (codegen_->GetCompilerOptions().IsBootImage()) {
+ if (codegen->GetCompilerOptions().IsBootImage()) {
// Compiling boot image. Resolve the string and allocate it if needed, to ensure
// the string will be added to the boot image.
DCHECK(!runtime->UseJitCompilation());
string = class_linker->ResolveString(dex_file, string_index, dex_cache);
CHECK(string != nullptr);
- if (compiler_driver_->GetSupportBootImageFixup()) {
- DCHECK(ContainsElement(compiler_driver_->GetDexFilesForOatFile(), &dex_file));
+ if (compiler_driver->GetSupportBootImageFixup()) {
+ DCHECK(ContainsElement(compiler_driver->GetDexFilesForOatFile(), &dex_file));
desired_load_kind = HLoadString::LoadKind::kBootImageLinkTimePcRelative;
} else {
// compiler_driver_test. Do not sharpen.
desired_load_kind = HLoadString::LoadKind::kRuntimeCall;
}
} else if (runtime->UseJitCompilation()) {
- DCHECK(!codegen_->GetCompilerOptions().GetCompilePic());
+ DCHECK(!codegen->GetCompilerOptions().GetCompilePic());
string = class_linker->LookupString(dex_file, string_index, dex_cache.Get());
if (string != nullptr) {
if (runtime->GetHeap()->ObjectIsInBootImageSpace(string)) {
@@ -283,7 +287,7 @@
// AOT app compilation. Try to lookup the string without allocating if not found.
string = class_linker->LookupString(dex_file, string_index, dex_cache.Get());
if (string != nullptr && runtime->GetHeap()->ObjectIsInBootImageSpace(string)) {
- if (codegen_->GetCompilerOptions().GetCompilePic()) {
+ if (codegen->GetCompilerOptions().GetCompilePic()) {
desired_load_kind = HLoadString::LoadKind::kBootImageInternTable;
} else {
desired_load_kind = HLoadString::LoadKind::kBootImageAddress;
@@ -293,12 +297,12 @@
}
}
if (string != nullptr) {
- load_string->SetString(handles_->NewHandle(string));
+ load_string->SetString(handles->NewHandle(string));
}
}
DCHECK_NE(desired_load_kind, static_cast<HLoadString::LoadKind>(-1));
- HLoadString::LoadKind load_kind = codegen_->GetSupportedLoadStringKind(desired_load_kind);
+ HLoadString::LoadKind load_kind = codegen->GetSupportedLoadStringKind(desired_load_kind);
load_string->SetLoadKind(load_kind);
}
diff --git a/compiler/optimizing/sharpening.h b/compiler/optimizing/sharpening.h
index bb1954e..6df7d6d 100644
--- a/compiler/optimizing/sharpening.h
+++ b/compiler/optimizing/sharpening.h
@@ -34,26 +34,29 @@
public:
HSharpening(HGraph* graph,
CodeGenerator* codegen,
- const DexCompilationUnit& compilation_unit,
CompilerDriver* compiler_driver,
- VariableSizedHandleScope* handles,
const char* name = kSharpeningPassName)
: HOptimization(graph, name),
codegen_(codegen),
- compilation_unit_(compilation_unit),
- compiler_driver_(compiler_driver),
- handles_(handles) { }
+ compiler_driver_(compiler_driver) { }
void Run() OVERRIDE;
static constexpr const char* kSharpeningPassName = "sharpening";
+ // Used by the builder.
+ static void ProcessLoadString(HLoadString* load_string,
+ CodeGenerator* codegen,
+ CompilerDriver* compiler_driver,
+ const DexCompilationUnit& dex_compilation_unit,
+ VariableSizedHandleScope* handles);
+
// Used by the builder and the inliner.
static HLoadClass::LoadKind ComputeLoadClassKind(HLoadClass* load_class,
CodeGenerator* codegen,
CompilerDriver* compiler_driver,
const DexCompilationUnit& dex_compilation_unit)
- REQUIRES_SHARED(Locks::mutator_lock_);
+ REQUIRES_SHARED(Locks::mutator_lock_);
// Used by Sharpening and InstructionSimplifier.
static void SharpenInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke,
@@ -61,12 +64,8 @@
CompilerDriver* compiler_driver);
private:
- void ProcessLoadString(HLoadString* load_string);
-
CodeGenerator* codegen_;
- const DexCompilationUnit& compilation_unit_;
CompilerDriver* compiler_driver_;
- VariableSizedHandleScope* handles_;
};
} // namespace art