Make object allocation entrypoints only take a class.
Change motivated by:
- Dex cache compression: having the allocation fast path do a
dex cache lookup will be too expensive. So instead, rely on the
compiler having direct access to the class (either through BSS for
AOT, or JIT tables for JIT).
- Inlining: the entrypoints relied on the caller of the allocation to
have the same dex cache as the outer method (stored at the bottom of
the stack). This meant we could not inline methods from a different
dex file that do allocations. By avoiding the dex cache lookup in
the entrypoint, we can now remove this restriction.
Code expansion on average for Docs/Gms/FB/Framework (go/lem numbers):
- Around 0.8% on arm64
- Around 1% for x64, arm
- Around 1.5% on x86
Test: test-art-host, test-art-target, ART_USE_READ_BARRIER=true/false
Test: test-art-host, test-art-target, ART_DEFAULT_GC_TYPE=SS ART_USE_TLAB=true
Change-Id: I41f3748bb4d251996aaf6a90fae4c50176f9295f
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index 8a7f6d3..1dd526f 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -3936,7 +3936,6 @@
} else {
InvokeRuntimeCallingConvention calling_convention;
locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
- locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
}
locations->SetOut(Location::RegisterLocation(R0));
}
@@ -3954,7 +3953,7 @@
codegen_->RecordPcInfo(instruction, instruction->GetDexPc());
} else {
codegen_->InvokeRuntime(instruction->GetEntrypoint(), instruction, instruction->GetDexPc());
- CheckEntrypointTypes<kQuickAllocObjectWithAccessCheck, void*, uint32_t, ArtMethod*>();
+ CheckEntrypointTypes<kQuickAllocObjectWithChecks, void*, mirror::Class*>();
}
}
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 5c33fe1..240e39d 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -4738,7 +4738,6 @@
locations->AddTemp(LocationFrom(kArtMethodRegister));
} else {
locations->SetInAt(0, LocationFrom(calling_convention.GetRegisterAt(0)));
- locations->SetInAt(1, LocationFrom(calling_convention.GetRegisterAt(1)));
}
locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimNot));
}
@@ -4756,7 +4755,7 @@
codegen_->RecordPcInfo(instruction, instruction->GetDexPc());
} else {
codegen_->InvokeRuntime(instruction->GetEntrypoint(), instruction, instruction->GetDexPc());
- CheckEntrypointTypes<kQuickAllocObjectWithAccessCheck, void*, uint32_t, ArtMethod*>();
+ CheckEntrypointTypes<kQuickAllocObjectWithChecks, void*, mirror::Class*>();
}
}
diff --git a/compiler/optimizing/code_generator_arm_vixl.cc b/compiler/optimizing/code_generator_arm_vixl.cc
index 00ad3e3..cf4d94d 100644
--- a/compiler/optimizing/code_generator_arm_vixl.cc
+++ b/compiler/optimizing/code_generator_arm_vixl.cc
@@ -3948,7 +3948,6 @@
} else {
InvokeRuntimeCallingConventionARMVIXL calling_convention;
locations->SetInAt(0, LocationFrom(calling_convention.GetRegisterAt(0)));
- locations->SetInAt(1, LocationFrom(calling_convention.GetRegisterAt(1)));
}
locations->SetOut(LocationFrom(r0));
}
@@ -3970,7 +3969,7 @@
codegen_->RecordPcInfo(instruction, instruction->GetDexPc());
} else {
codegen_->InvokeRuntime(instruction->GetEntrypoint(), instruction, instruction->GetDexPc());
- CheckEntrypointTypes<kQuickAllocObjectWithAccessCheck, void*, uint32_t, ArtMethod*>();
+ CheckEntrypointTypes<kQuickAllocObjectWithChecks, void*, mirror::Class*>();
}
}
diff --git a/compiler/optimizing/code_generator_mips.cc b/compiler/optimizing/code_generator_mips.cc
index 01e0dac..29f8b2a 100644
--- a/compiler/optimizing/code_generator_mips.cc
+++ b/compiler/optimizing/code_generator_mips.cc
@@ -5900,7 +5900,6 @@
locations->AddTemp(Location::RegisterLocation(kMethodRegisterArgument));
} else {
locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
- locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
}
locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimNot));
}
@@ -5917,7 +5916,7 @@
codegen_->RecordPcInfo(instruction, instruction->GetDexPc());
} else {
codegen_->InvokeRuntime(instruction->GetEntrypoint(), instruction, instruction->GetDexPc());
- CheckEntrypointTypes<kQuickAllocObjectWithAccessCheck, void*, uint32_t, ArtMethod*>();
+ CheckEntrypointTypes<kQuickAllocObjectWithChecks, void*, mirror::Class*>();
}
}
diff --git a/compiler/optimizing/code_generator_mips64.cc b/compiler/optimizing/code_generator_mips64.cc
index 36690c0..dd3f0fe 100644
--- a/compiler/optimizing/code_generator_mips64.cc
+++ b/compiler/optimizing/code_generator_mips64.cc
@@ -3841,7 +3841,6 @@
locations->AddTemp(Location::RegisterLocation(kMethodRegisterArgument));
} else {
locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
- locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
}
locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimNot));
}
@@ -3859,7 +3858,7 @@
codegen_->RecordPcInfo(instruction, instruction->GetDexPc());
} else {
codegen_->InvokeRuntime(instruction->GetEntrypoint(), instruction, instruction->GetDexPc());
- CheckEntrypointTypes<kQuickAllocObjectWithAccessCheck, void*, uint32_t, ArtMethod*>();
+ CheckEntrypointTypes<kQuickAllocObjectWithChecks, void*, mirror::Class*>();
}
}
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 0abe855..786bc50 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -4150,7 +4150,6 @@
} else {
InvokeRuntimeCallingConvention calling_convention;
locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
- locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
}
}
@@ -4166,7 +4165,7 @@
codegen_->RecordPcInfo(instruction, instruction->GetDexPc());
} else {
codegen_->InvokeRuntime(instruction->GetEntrypoint(), instruction, instruction->GetDexPc());
- CheckEntrypointTypes<kQuickAllocObjectWithAccessCheck, void*, uint32_t, ArtMethod*>();
+ CheckEntrypointTypes<kQuickAllocObjectWithChecks, void*, mirror::Class*>();
DCHECK(!codegen_->IsLeafMethod());
}
}
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 903844f..06b48c4 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -4038,7 +4038,6 @@
locations->AddTemp(Location::RegisterLocation(kMethodRegisterArgument));
} else {
locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
- locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
}
locations->SetOut(Location::RegisterLocation(RAX));
}
@@ -4055,7 +4054,7 @@
codegen_->RecordPcInfo(instruction, instruction->GetDexPc());
} else {
codegen_->InvokeRuntime(instruction->GetEntrypoint(), instruction, instruction->GetDexPc());
- CheckEntrypointTypes<kQuickAllocObjectWithAccessCheck, void*, uint32_t, ArtMethod*>();
+ CheckEntrypointTypes<kQuickAllocObjectWithChecks, void*, mirror::Class*>();
DCHECK(!codegen_->IsLeafMethod());
}
}
diff --git a/compiler/optimizing/inliner.cc b/compiler/optimizing/inliner.cc
index e5d05e9..c970e5c 100644
--- a/compiler/optimizing/inliner.cc
+++ b/compiler/optimizing/inliner.cc
@@ -1428,15 +1428,6 @@
return false;
}
- if (current->IsNewInstance() &&
- (current->AsNewInstance()->GetEntrypoint() == kQuickAllocObjectWithAccessCheck)) {
- VLOG(compiler) << "Method " << callee_dex_file.PrettyMethod(method_index)
- << " could not be inlined because it is using an entrypoint"
- << " with access checks";
- // Allocation entrypoint does not handle inlined frames.
- return false;
- }
-
if (current->IsNewArray() &&
(current->AsNewArray()->GetEntrypoint() == kQuickAllocArrayWithAccessCheck)) {
VLOG(compiler) << "Method " << callee_dex_file.PrettyMethod(method_index)
diff --git a/compiler/optimizing/instruction_builder.cc b/compiler/optimizing/instruction_builder.cc
index 768b1d8..009d549 100644
--- a/compiler/optimizing/instruction_builder.cc
+++ b/compiler/optimizing/instruction_builder.cc
@@ -917,11 +917,11 @@
bool finalizable;
bool needs_access_check = NeedsAccessCheck(type_index, dex_cache, &finalizable);
- // Only the non-resolved entrypoint handles the finalizable class case. If we
+ // Only the access check entrypoint handles the finalizable class case. If we
// need access checks, then we haven't resolved the method and the class may
// again be finalizable.
QuickEntrypointEnum entrypoint = (finalizable || needs_access_check)
- ? kQuickAllocObject
+ ? kQuickAllocObjectWithChecks
: kQuickAllocObjectInitialized;
if (outer_dex_cache.Get() != dex_cache.Get()) {
@@ -946,7 +946,6 @@
AppendInstruction(new (arena_) HNewInstance(
cls,
- graph_->GetCurrentMethod(),
dex_pc,
type_index,
*dex_compilation_unit_->GetDexFile(),
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index e1e0b7d..db1b277 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -3774,10 +3774,9 @@
DISALLOW_COPY_AND_ASSIGN(HCompare);
};
-class HNewInstance FINAL : public HExpression<2> {
+class HNewInstance FINAL : public HExpression<1> {
public:
HNewInstance(HInstruction* cls,
- HCurrentMethod* current_method,
uint32_t dex_pc,
dex::TypeIndex type_index,
const DexFile& dex_file,
@@ -3791,7 +3790,6 @@
SetPackedFlag<kFlagNeedsAccessCheck>(needs_access_check);
SetPackedFlag<kFlagFinalizable>(finalizable);
SetRawInputAt(0, cls);
- SetRawInputAt(1, current_method);
}
dex::TypeIndex GetTypeIndex() const { return type_index_; }
diff --git a/compiler/optimizing/prepare_for_register_allocation.cc b/compiler/optimizing/prepare_for_register_allocation.cc
index f9ac3a0..db7c1fb 100644
--- a/compiler/optimizing/prepare_for_register_allocation.cc
+++ b/compiler/optimizing/prepare_for_register_allocation.cc
@@ -134,39 +134,6 @@
}
}
-void PrepareForRegisterAllocation::VisitNewInstance(HNewInstance* instruction) {
- HLoadClass* load_class = instruction->InputAt(0)->AsLoadClass();
- const bool has_only_one_use = load_class->HasOnlyOneNonEnvironmentUse();
- // Change the entrypoint to kQuickAllocObject if either:
- // - the class is finalizable (only kQuickAllocObject handles finalizable classes),
- // - the class needs access checks (we do not know if it's finalizable),
- // - or the load class has only one use.
- if (instruction->IsFinalizable() || has_only_one_use || load_class->NeedsAccessCheck()) {
- instruction->SetEntrypoint(kQuickAllocObject);
- instruction->ReplaceInput(GetGraph()->GetIntConstant(load_class->GetTypeIndex().index_), 0);
- if (has_only_one_use) {
- // We've just removed the only use of the HLoadClass. Since we don't run DCE after this pass,
- // do it manually if possible.
- if (!load_class->CanThrow()) {
- // If the load class can not throw, it has no side effects and can be removed if there is
- // only one use.
- load_class->GetBlock()->RemoveInstruction(load_class);
- } else if (!instruction->GetEnvironment()->IsFromInlinedInvoke() &&
- CanMoveClinitCheck(load_class, instruction)) {
- // The allocation entry point that deals with access checks does not work with inlined
- // methods, so we need to check whether this allocation comes from an inlined method.
- // We also need to make the same check as for moving clinit check, whether the HLoadClass
- // has the clinit check responsibility or not (HLoadClass can throw anyway).
- // If it needed access checks, we delegate the access check to the allocation.
- if (load_class->NeedsAccessCheck()) {
- instruction->SetEntrypoint(kQuickAllocObjectWithAccessCheck);
- }
- load_class->GetBlock()->RemoveInstruction(load_class);
- }
- }
- }
-}
-
bool PrepareForRegisterAllocation::CanEmitConditionAt(HCondition* condition,
HInstruction* user) const {
if (condition->GetNext() != user) {
diff --git a/compiler/optimizing/prepare_for_register_allocation.h b/compiler/optimizing/prepare_for_register_allocation.h
index a679148..c128227 100644
--- a/compiler/optimizing/prepare_for_register_allocation.h
+++ b/compiler/optimizing/prepare_for_register_allocation.h
@@ -44,7 +44,6 @@
void VisitClinitCheck(HClinitCheck* check) OVERRIDE;
void VisitCondition(HCondition* condition) OVERRIDE;
void VisitInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) OVERRIDE;
- void VisitNewInstance(HNewInstance* instruction) OVERRIDE;
bool CanMoveClinitCheck(HInstruction* input, HInstruction* user) const;
bool CanEmitConditionAt(HCondition* condition, HInstruction* user) const;