Change suspend entrypoint to save all registers.

We avoid the need to save/restore registers in slow paths
and get significant code size savings. On Nexus 9, AOSP:
  - 32-bit boot.oat: -1.4MiB (-1.9%)
  - 64-bit boot.oat: -2.0MiB (-2.3%)
  - other 32-bit oat files in dalvik-cache: -200KiB (-1.7%)
  - other 64-bit oat files in dalvik-cache: -2.3MiB (-2.1%)

Test: Run ART test suite on host and Nexus 9 with gc stress.
Bug: 30212852
Change-Id: I7015afc1e7d30341618c9200a3dc9ae277afd134

8 files changed, 19 insertions, 21 deletions

diff --git a/compiler/image_writer.cc b/compiler/image_writer.cc
index 063eb11718..0f6944ac6a 100644
--- a/compiler/image_writer.cc
+++ b/compiler/image_writer.cc
@@ -1377,6 +1377,8 @@ void ImageWriter::CalculateNewObjectOffsets() {
       runtime->GetCalleeSaveMethod(Runtime::kRefsOnly);
   image_methods_[ImageHeader::kRefsAndArgsSaveMethod] =
       runtime->GetCalleeSaveMethod(Runtime::kRefsAndArgs);
+  image_methods_[ImageHeader::kSaveEverythingMethod] =
+      runtime->GetCalleeSaveMethod(Runtime::kSaveEverything);
   // Visit image methods first to have the main runtime methods in the first image.
   for (auto* m : image_methods_) {
     CHECK(m != nullptr);
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index 3269dc6605..94ecb1c63c 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -765,16 +765,24 @@ void CodeGenerator::RecordPcInfo(HInstruction* instruction,
   LocationSummary* locations = instruction->GetLocations();
 
   uint32_t register_mask = locations->GetRegisterMask();
-  if (locations->OnlyCallsOnSlowPath()) {
-    // In case of slow path, we currently set the location of caller-save registers
-    // to register (instead of their stack location when pushed before the slow-path
-    // call). Therefore register_mask contains both callee-save and caller-save
-    // registers that hold objects. We must remove the caller-save from the mask, since
-    // they will be overwritten by the callee.
-    register_mask &= core_callee_save_mask_;
+  if (instruction->IsSuspendCheck()) {
+    // Suspend check has special ABI that saves the caller-save registers in callee,
+    // so we want to emit stack maps containing the registers.
+    // TODO: Register allocator still reserves space for the caller-save registers.
+    // We should add slow-path-specific caller-save information into LocationSummary
+    // and refactor the code here as well as in the register allocator to use it.
+  } else {
+    if (locations->OnlyCallsOnSlowPath()) {
+      // In case of slow path, we currently set the location of caller-save registers
+      // to register (instead of their stack location when pushed before the slow-path
+      // call). Therefore register_mask contains both callee-save and caller-save
+      // registers that hold objects. We must remove the caller-save from the mask, since
+      // they will be overwritten by the callee.
+      register_mask &= core_callee_save_mask_;
+    }
+    // The register mask must be a subset of callee-save registers.
+    DCHECK_EQ(register_mask & core_callee_save_mask_, register_mask);
   }
-  // The register mask must be a subset of callee-save registers.
-  DCHECK_EQ(register_mask & core_callee_save_mask_, register_mask);
   stack_map_stream_.BeginStackMapEntry(outer_dex_pc,
                                        native_pc,
                                        register_mask,
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index 124a61fe67..62f3ee9be7 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -119,11 +119,9 @@ class SuspendCheckSlowPathARM : public SlowPathCode {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorARM* arm_codegen = down_cast<CodeGeneratorARM*>(codegen);
     __ Bind(GetEntryLabel());
-    SaveLiveRegisters(codegen, instruction_->GetLocations());
     arm_codegen->InvokeRuntime(
         QUICK_ENTRY_POINT(pTestSuspend), instruction_, instruction_->GetDexPc(), this);
     CheckEntrypointTypes<kQuickTestSuspend, void, void>();
-    RestoreLiveRegisters(codegen, instruction_->GetLocations());
     if (successor_ == nullptr) {
       __ b(GetReturnLabel());
     } else {
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index efeef7bb6f..21ca506077 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -398,11 +398,9 @@ class SuspendCheckSlowPathARM64 : public SlowPathCodeARM64 {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorARM64* arm64_codegen = down_cast<CodeGeneratorARM64*>(codegen);
     __ Bind(GetEntryLabel());
-    SaveLiveRegisters(codegen, instruction_->GetLocations());
     arm64_codegen->InvokeRuntime(
         QUICK_ENTRY_POINT(pTestSuspend), instruction_, instruction_->GetDexPc(), this);
     CheckEntrypointTypes<kQuickTestSuspend, void, void>();
-    RestoreLiveRegisters(codegen, instruction_->GetLocations());
     if (successor_ == nullptr) {
       __ B(GetReturnLabel());
     } else {
diff --git a/compiler/optimizing/code_generator_mips.cc b/compiler/optimizing/code_generator_mips.cc
index 334d30d90e..bdd75b634b 100644
--- a/compiler/optimizing/code_generator_mips.cc
+++ b/compiler/optimizing/code_generator_mips.cc
@@ -351,14 +351,12 @@ class SuspendCheckSlowPathMIPS : public SlowPathCodeMIPS {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorMIPS* mips_codegen = down_cast<CodeGeneratorMIPS*>(codegen);
     __ Bind(GetEntryLabel());
-    SaveLiveRegisters(codegen, instruction_->GetLocations());
     mips_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pTestSuspend),
                                 instruction_,
                                 instruction_->GetDexPc(),
                                 this,
                                 IsDirectEntrypoint(kQuickTestSuspend));
     CheckEntrypointTypes<kQuickTestSuspend, void, void>();
-    RestoreLiveRegisters(codegen, instruction_->GetLocations());
     if (successor_ == nullptr) {
       __ B(GetReturnLabel());
     } else {
diff --git a/compiler/optimizing/code_generator_mips64.cc b/compiler/optimizing/code_generator_mips64.cc
index 29b8c206e6..e16abc52d1 100644
--- a/compiler/optimizing/code_generator_mips64.cc
+++ b/compiler/optimizing/code_generator_mips64.cc
@@ -300,13 +300,11 @@ class SuspendCheckSlowPathMIPS64 : public SlowPathCodeMIPS64 {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
     __ Bind(GetEntryLabel());
-    SaveLiveRegisters(codegen, instruction_->GetLocations());
     mips64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pTestSuspend),
                                   instruction_,
                                   instruction_->GetDexPc(),
                                   this);
     CheckEntrypointTypes<kQuickTestSuspend, void, void>();
-    RestoreLiveRegisters(codegen, instruction_->GetLocations());
     if (successor_ == nullptr) {
       __ Bc(GetReturnLabel());
     } else {
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 528e94f76b..3fb4fe8c72 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -192,13 +192,11 @@ class SuspendCheckSlowPathX86 : public SlowPathCode {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorX86* x86_codegen = down_cast<CodeGeneratorX86*>(codegen);
     __ Bind(GetEntryLabel());
-    SaveLiveRegisters(codegen, instruction_->GetLocations());
     x86_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pTestSuspend),
                                instruction_,
                                instruction_->GetDexPc(),
                                this);
     CheckEntrypointTypes<kQuickTestSuspend, void, void>();
-    RestoreLiveRegisters(codegen, instruction_->GetLocations());
     if (successor_ == nullptr) {
       __ jmp(GetReturnLabel());
     } else {
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 0f0129b421..b0c3caa5b1 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -149,13 +149,11 @@ class SuspendCheckSlowPathX86_64 : public SlowPathCode {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorX86_64* x86_64_codegen = down_cast<CodeGeneratorX86_64*>(codegen);
     __ Bind(GetEntryLabel());
-    SaveLiveRegisters(codegen, instruction_->GetLocations());
     x86_64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pTestSuspend),
                                   instruction_,
                                   instruction_->GetDexPc(),
                                   this);
     CheckEntrypointTypes<kQuickTestSuspend, void, void>();
-    RestoreLiveRegisters(codegen, instruction_->GetLocations());
     if (successor_ == nullptr) {
       __ jmp(GetReturnLabel());
     } else {