5 files changed, 16 insertions, 28 deletions

diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index 47e6625d07..5e6e175c67 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -80,7 +80,11 @@ class SlowPathCode : public DeletableArenaObject<kArenaAllocSlowPaths> {
 
   virtual void EmitNativeCode(CodeGenerator* codegen) = 0;
 
+  // Save live core and floating-point caller-save registers and
+  // update the stack mask in `locations` for registers holding object
+  // references.
   virtual void SaveLiveRegisters(CodeGenerator* codegen, LocationSummary* locations);
+  // Restore live core and floating-point caller-save registers.
   virtual void RestoreLiveRegisters(CodeGenerator* codegen, LocationSummary* locations);
 
   bool IsCoreRegisterSaved(int reg) const {
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index 1aa7b5404c..474e9d540a 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -437,11 +437,9 @@ class ReadBarrierMarkSlowPathARM : public SlowPathCode {
         << instruction_->DebugName();
 
     __ Bind(GetEntryLabel());
-    // Save live registers before the runtime call, and in particular
-    // R0 (if it is live), as it is clobbered by functions
-    // art_quick_read_barrier_mark_regX.
-    SaveLiveRegisters(codegen, locations);
-
+    // No need to save live registers; it's taken care of by the
+    // entrypoint. Also, there is no need to update the stack mask,
+    // as this runtime call will not trigger a garbage collection.
     InvokeRuntimeCallingConvention calling_convention;
     CodeGeneratorARM* arm_codegen = down_cast<CodeGeneratorARM*>(codegen);
     DCHECK_NE(reg, SP);
@@ -469,8 +467,6 @@ class ReadBarrierMarkSlowPathARM : public SlowPathCode {
                                instruction_,
                                instruction_->GetDexPc(),
                                this);
-
-    RestoreLiveRegisters(codegen, locations);
     __ b(GetExitLabel());
   }
 
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 07b7823571..cec641f27c 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -603,11 +603,9 @@ class ReadBarrierMarkSlowPathARM64 : public SlowPathCodeARM64 {
         << instruction_->DebugName();
 
     __ Bind(GetEntryLabel());
-    // Save live registers before the runtime call, and in particular
-    // W0 (if it is live), as it is clobbered by functions
-    // art_quick_read_barrier_mark_regX.
-    SaveLiveRegisters(codegen, locations);
-
+    // No need to save live registers; it's taken care of by the
+    // entrypoint. Also, there is no need to update the stack mask,
+    // as this runtime call will not trigger a garbage collection.
     InvokeRuntimeCallingConvention calling_convention;
     CodeGeneratorARM64* arm64_codegen = down_cast<CodeGeneratorARM64*>(codegen);
     DCHECK_NE(obj_.reg(), LR);
@@ -635,8 +633,6 @@ class ReadBarrierMarkSlowPathARM64 : public SlowPathCodeARM64 {
                                  instruction_,
                                  instruction_->GetDexPc(),
                                  this);
-
-    RestoreLiveRegisters(codegen, locations);
     __ B(GetExitLabel());
   }
 
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 1cc6060f68..93bf022a32 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -472,11 +472,9 @@ class ReadBarrierMarkSlowPathX86 : public SlowPathCode {
         << instruction_->DebugName();
 
     __ Bind(GetEntryLabel());
-    // Save live registers before the runtime call, and in particular
-    // EAX (if it is live), as it is clobbered by functions
-    // art_quick_read_barrier_mark_regX.
-    SaveLiveRegisters(codegen, locations);
-
+    // No need to save live registers; it's taken care of by the
+    // entrypoint. Also, there is no need to update the stack mask,
+    // as this runtime call will not trigger a garbage collection.
     InvokeRuntimeCallingConvention calling_convention;
     CodeGeneratorX86* x86_codegen = down_cast<CodeGeneratorX86*>(codegen);
     DCHECK_NE(reg, ESP);
@@ -502,8 +500,6 @@ class ReadBarrierMarkSlowPathX86 : public SlowPathCode {
                                instruction_,
                                instruction_->GetDexPc(),
                                this);
-
-    RestoreLiveRegisters(codegen, locations);
     __ jmp(GetExitLabel());
   }
 
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index a0158938b5..0d85bea8c0 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -493,11 +493,9 @@ class ReadBarrierMarkSlowPathX86_64 : public SlowPathCode {
         << instruction_->DebugName();
 
     __ Bind(GetEntryLabel());
-    // Save live registers before the runtime call, and in particular
-    // RDI and/or RAX (if they are live), as they are clobbered by
-    // functions art_quick_read_barrier_mark_regX.
-    SaveLiveRegisters(codegen, locations);
-
+    // No need to save live registers; it's taken care of by the
+    // entrypoint. Also, there is no need to update the stack mask,
+    // as this runtime call will not trigger a garbage collection.
     InvokeRuntimeCallingConvention calling_convention;
     CodeGeneratorX86_64* x86_64_codegen = down_cast<CodeGeneratorX86_64*>(codegen);
     DCHECK_NE(reg, RSP);
@@ -523,8 +521,6 @@ class ReadBarrierMarkSlowPathX86_64 : public SlowPathCode {
                                   instruction_,
                                   instruction_->GetDexPc(),
                                   this);
-
-    RestoreLiveRegisters(codegen, locations);
     __ jmp(GetExitLabel());
   }